NVIDIA / NVIDIA/nvcf

feat(stargate): reject unsupported server identities in secure mode

Open
#946 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Go
Stars
218
Forks
72
Avg merge
1d 9h
Merged PRs (30d)
417

Description

Problem

Stargate generates a self-signed server identity when both TLS certificate and
private key inputs are absent. Setting quic_insecure=false controls outbound
certificate verification, but it does not disable this inbound server fallback.

Self-managed secure PKI deployments need an explicit fail-closed server identity
policy. Missing, implicit, or explicitly self-signed service identities must not
start a secure reverse QUIC listener.

Required behavior

  • Add explicit development and strict server identity policies.
  • Preserve generated self-signed identities only in explicit development mode.
  • Require both certificate and private key material in strict mode.
  • Reject malformed material, a mismatched private key, and an empty certificate
    chain.
  • Reject an explicitly self-signed service leaf in strict mode.
  • Reject a CA certificate used as the service leaf.
  • Reject an expired, not-yet-valid, or non-server certificate.
  • Validate the service leaf against the effective advertised hostname.
  • Fail before opening the reverse QUIC listener when strict validation fails.
  • Do not fall back to generated identity or insecure verification after failure.

Chart integration

  • Expose the policy through the LLM request-router chart.
  • Select strict policy for the self-managed secure PKI profile.
  • Keep existing development behavior for non-secure development installs.
  • Apply the same strict validation to cert-manager identities and any future
    existing-Secret identity source.

Acceptance criteria

  • Strict mode with missing certificate or key material fails startup.
  • Strict mode rejects a generated or explicitly supplied self-signed service
    leaf.
  • Strict mode accepts a valid CA-issued service identity with a matching key and
    advertised hostname.
  • The request-router does not become ready after strict identity validation
    fails.
  • Unit and chart tests cover strict and development behavior.
  • End-to-end validation records the valid and fail-closed cases.

Related

Relates to #19.

Follow-up to #51.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start at Stargate's reverse QUIC listener and server-identity generation, then inspect the LLM request-router chart and its cert-manager identity integration. Done means strict and development policies behave as specified, invalid identities fail before listener readiness, and unit, chart, and end-to-end tests cover valid and fail-closed cases.

Written by the indexing model from the issue text.

Assessment

Tech stack
go, helm, kubernetes
Domain
backend, cloud, networking, security
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.