NVIDIA / NVIDIA/nvcf

test(bdd): prove three-replica Stargate registration in the multi-cluster Helmfile flow

Open
#1,420 0 comments 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

Why

The secure multi-cluster registration BDD is intended to prove that a Pylon registers with every LLM request-router replica. It asserts router identities 0, 1, and 2 and three registration/tunnel connections, but the shared local fixture currently configures addons.llm.requestRouter.replicaCount: 1.

Make the three-replica topology explicit so the live BDD proves the supported minimum replica configuration rather than relying on fixture defaults or fake wiring output.

Scope

Extend tests/bdd/features/multi-cluster-helmfile-llm-registration-tls.feature.

  • Set addons.llm.requestRouter.replicaCount: 3 in the feature-specific control-plane environment values.
  • Keep DNS discovery enabled for the multi-replica StatefulSet.
  • After installation, use an operator-visible kubectl command to assert that llm-request-router has desired replicas 3 and ready replicas 3.
  • Retain the trusted Watch assertion for router identities llm-request-router-0, -1, and -2.
  • Retain the Pylon metrics assertions that prove three registration streams and three reverse tunnels.
  • Retain the authenticated LLM invocation after registration.
  • Update the focused Godog wiring test for the new StatefulSet observation.

Acceptance criteria

  • The live feature explicitly configures three LLM request-router replicas.
  • The control-plane StatefulSet reports three desired and three ready replicas.
  • A compute-plane Pylon is observable as connected to all three replicas.
  • An authenticated LLM request succeeds after all three registrations are established.
  • The feature fails when the configured replica count or a router registration is missing.

Non-goals

  • Do not change the shared local BDD fixture default replica count.
  • Do not add duplicate three-replica lifecycle coverage to generic Helmfile install smoke tests.
  • Do not test router failover or pod deletion in this issue.

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 tests/bdd/features/multi-cluster-helmfile-llm-registration-tls.feature and the focused Godog wiring test. Set the feature-specific request-router replica count to 3, keep DNS discovery and the existing Watch, metrics, tunnel, and authenticated LLM assertions, and add the operator-visible StatefulSet check. Done means the live BDD observes three desired and ready replicas and fails if any registration or the final request is missing.

Written by the indexing model from the issue text.

Assessment

Tech stack
go, helm, kubernetes
Domain
devops, infrastructure, testing-qa
Issue type
Feature
Difficulty
3/5
Estimated time
1-2 days
Activity status
Active
Clarity
Clearly specified
Newbie friendliness
74/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.