If you discover a security issue in l9gpu, please report it privately rather than opening a public GitHub issue.

Email: support@last9.io

Please include:

• A description of the issue and its impact
• Steps to reproduce
• Affected version(s) and component (l9gpu, k8sprocessor, slurmprocessor, k8shelper, shelper, Helm chart, alerts)
• Any suggested mitigation

• We will acknowledge receipt within 3 business days.
• We aim to provide an initial assessment within 10 business days.
• Coordinated disclosure: we request a 90-day embargo from report date to public disclosure, extendable by mutual agreement for complex issues.

In scope:

• Code in this repository (Python package, Go modules, Helm chart, deployment manifests)
• Official container images published from this repo

Out of scope:

• Third-party dependencies (please report upstream; we will bump when a patch lands)
• Vulnerabilities requiring physical access to GPU nodes or root on the host

Thank you for helping keep the project and its users safe.

Python installations require certifi>=2026.10.19. The development lockfile also includes patched pip>=26.2.2, which carries its own CA bundle. Go modules and collector builds require Go 1.27.3 or newer. See the incident and patched-version list and the Go security announcement.

After installing the locked dependencies, run this network smoke check:

python scripts/check_tls_trust.py

It must accept a trusted HTTPS endpoint and reject both the direct and cross-signed compromised test chains with certificate verification errors. Connection failures do not count as successful rejection. Add --system to check the host or container system CA bundle too.

Rebuild Python and collector release artifacts with the updated dependencies and Go compiler before publishing a new release. Rebuild container images with --pull --no-cache to refresh base-image CA bundles and dependency layers; existing images and installed packages do not receive these changes automatically. The Python image refreshes Debian's ca-certificates package in both stages. Verify a rebuilt Python image before publishing:

docker build --pull --no-cache -f docker/Dockerfile -t l9gpu:tls-check .
docker run --rm --entrypoint python \
  -v "$PWD/scripts:/checks:ro" l9gpu:tls-check \
  /checks/check_tls_trust.py --system

Publish rebuilt packages and images through the normal release workflows. Do not disable certificate verification to restore connectivity.