skypilot-org/skypilot

[Kubernetes] GPU prefix matching should not allow non-existent GPU variants

Open

#9035 opened on Mar 10, 2026

View on GitHub
 (2 comments) (0 reactions) (0 assignees)Python (4,859 stars) (311 forks)batch import
good first issuegood starter issues

Description

On a K8s cluster where sky show-gpus --infra k8s shows only H100:

  • sky launch --gpus H100-MEGA succeeds (should it?)
  • sky launch --gpus 'NVIDIA:40GB+' shows H100-NVLINK, H100-SXM, H100-MEGA, H100-NVL as separate options, all resolving to the same physical H100

K8s prefix-based fuzzy matching in _accelerator_name_matches() accepts any name where H100 is a prefix (e.g., H100-MEGA, H100-SXM). The optimizer should filter against what's actually available on the cluster.

Contributor guide

Tech stack
pythonkubernetes
Domain
backendinfrastructurecloud
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
fresh
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
Pythonbasic Kubernetes GPU concepts
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60
Research direction
Locate the function ` accelerator name matches()` in the SkyPilot codebase that performs prefix based fuzzy matching for GPU names on Kubernetes. The function currently accepts any name where the actual GPU name (e.g., H100) is a prefix. Modify it to filter against the set of GPU variants actually available on the cluster, as reported by `sky show gpus infra k8s`. Consider using exact matching or look up the available variants before deciding whether a requested GPU variant is valid.