NVIDIA / NVIDIA/nvcf

Improve WaitAndWiden affinity fallback behavior

Open
#1,659 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

Problem

WaitAndWiden treats cache-affinity candidates as unordered peers. A load comparator can move a session between its primary and backup while the primary is still admissible, which weakens cache locality. Queue admissibility also determines when global candidates enter selection, so locality can be abandoned before the request SLO requires it.

Desired behavior

  • Treat cache-affinity candidates as an ordered primary and backup.
  • Use the backup only when the primary is not admissible.
  • Unlock non-affinity candidates only after both affinity candidates are unsuitable.
  • Derive the affinity wait budget from the request SLO while leaving enough time for the best cold candidate.

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 by locating the Go implementation of WaitAndWiden and trace how affinity candidates, queue admissibility, and global candidates enter selection. Verify the primary remains preferred, the backup is used only when needed, and non-affinity candidates wait until both are unsuitable. Confirm the affinity wait budget follows the request SLO while preserving time for the best cold candidate.

Written by the indexing model from the issue text.

Assessment

Tech stack
go
Domain
backend-api-design
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
48/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.