NVIDIA / NVIDIA/nvcf

Make ICMS instance-placement arbitration race-safe with NATS

Open
#43 0 comments 0 reactions 1 assignee View on GitHub

@dmikhaylovnv is already working on this.

Since Jul 15, 2026.

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

Description

Description

Make ICMS instance-placement arbitration consistent and race-safe by using NATS for request broadcast and a request-scoped JetStream consumer to collect cluster acknowledgements. Preserve the existing auction model while ensuring one ICMS instance owns each decision.

Definition of Done

  • Instance-creation requests can target the intended region or set of clusters over NATS.
  • Acknowledgements for each request are collected by a short-lived JetStream consumer owned by the publishing ICMS instance.
  • Exactly one eligible cluster is selected and every responder receives a deterministic winning or non-winning result.
  • No-responder and publisher-failure paths use bounded timeouts and retry-safe behavior.
  • The selected result is persisted so retries can resume without repeating a completed arbitration.
  • The new path can run without side effects for comparison with the current flow before a feature-flagged cutover.

Out of Scope

Replacing messaging for unrelated components is not part of 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.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.