NVIDIA / NVIDIA/nvcf

fix(llm-routing): consume scheme-aware Pylon and NVCA releases

Open
#1,362 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Go
Stars
218
Forks
72
Avg merge
1d 12h
Merged PRs (30d)
427

Description

Problem

The self-managed stack uses a plaintext in-cluster Stargate registration endpoint and authenticated reverse QUIC. Current Pylon and NVCA releases can apply the reverse-QUIC trust bundle as a gRPC certificate authority even when the registration endpoint uses explicit HTTP. Worker registration then fails.

Issue #1367 tracks the paired source fix. Both components need released artifacts before the stack can consume the corrected contract. PR #1363 currently carries a temporary Pylon 0.13.0 pin only to keep the new smoke coverage running while these fixes are separated.

Scope

  • Publish a Pylon release containing the scheme-aware gRPC CA selection fix from #1367.
  • Publish an NVCA release that only injects a gRPC CA for HTTPS Stargate addresses.
  • Update the self-managed control-plane Pylon version.
  • Update the compute-plane NVCA version.
  • Remove the temporary Pylon downgrade from #1363.
  • Update the version catalog and generated artifact manifest.
  • Rerun the self-managed LLM, autoscaler, and Vanity Gateway smoke coverage.

Definition of done

  • Fixed Pylon and NVCA images are available from the supported registry.
  • Plaintext in-cluster registration works without a gRPC CA override.
  • HTTPS registration still uses the configured trust bundle.
  • Reverse QUIC continues to authenticate with its transport trust bundle.
  • The control-plane and compute-plane stacks consume the fixed releases.
  • Documentation and the generated manifest report the fixed versions.
  • Focused component tests and self-managed BDD smoke tests pass.

Blocked by #1367.
Related to #1361.

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 with blocked issue #1367 and the temporary Pylon pin in PR #1363 to understand the separated fixes. Update the version catalog, generated artifact manifest, control-plane Pylon version, and compute-plane NVCA version after fixed releases are available. Run the focused component tests and self-managed LLM, autoscaler, and Vanity Gateway smoke coverage, confirming plaintext and HTTPS registration behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
go, grpc
Domain
infrastructure, release, testing
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Clearly specified
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.