NVIDIA / NVIDIA/nvcf

Model cache fast-follows: sharedfs data-sharing probe, Samba infra reconciliation, ephemeral env filtering

Open
#250 0 comments 0 reactions 1 assignee View on GitHub

@balajinvda is already working on this.

Since Jul 20, 2026.

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

Description

Follow-ups deferred from #227 (also listed in the SDD Known gaps section):

  1. sharedfs reader binding assumes the StorageClass exposes one shared filesystem. Separately provisioned PVCs do not share data on provisioners that create per-claim access points, subvolumes, or directories (EFS access points, CephFS subvolumes, some NFS provisioners). The capability probe validates bindability, not data-sharing. Fast-follow: a write-through-one-claim / read-through-another probe, and deriving reader PVs from the writer's bound volume where the driver supports it.

  2. The per-handle Samba infrastructure (EnsureSambaModelCacheInfra) is create-once: image, resource, and cache-size changes do not reconcile onto an existing server, and the backing PVC is never expanded for a larger cache-size request. Fast-follow: reconcile mutable Deployment fields and expand the PVC, keeping credentials create-once.

  3. The ephemeral model-cache-init env forwards the full launch environment through the miniservice metadata ConfigMap. Fast-follow: narrow it to the variables the init container consumes and source sensitive values from Secrets.

Relates to #226

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.