Skip to content

docs: add remote backend Valkey option for KV offloading + Feast#703

Open
edlng wants to merge 1 commit into
kserve:mainfrom
edlng:valkey-kvcache-offloading-docs
Open

docs: add remote backend Valkey option for KV offloading + Feast#703
edlng wants to merge 1 commit into
kserve:mainfrom
edlng:valkey-kvcache-offloading-docs

Conversation

@edlng

@edlng edlng commented Jul 15, 2026

Copy link
Copy Markdown

Summary

Adds a "Using Valkey as Remote Backend" section to the KV-cache offloading guide (right next to the existing Redis section) and a "Using Valkey instead of Redis" section to the Feast feature store guide. Both changes are mirrored into versioned_docs/version-0.18/ so the current and 0.18 docs stay in sync.

Valkey already works as a drop-in Redis replacement for both of these: LMCache's ValkeyConnector speaks the same wire protocol, and Feast's Redis online-store client doesn't care whether it's talking to Redis or Valkey on the other end.

Issue

Related: #701

Changes

  • docs/model-serving/generative-inference/kvcache-offloading/kvcache-offloading.md -- new Valkey section with ConfigMap/Deployment/Service YAML, plus a link to the paired sample manifest in kserve/kserve.
  • docs/model-serving/predictive-inference/transformers/feast-feature-store/feast-feature-store.md -- new section pointing at the valkey.yaml sample and noting the one thing that needs manual attention: the feature-store-initializer entrypoint script hardcodes redis-service in its connection string, so anyone using the Valkey sample needs to update that before building the initializer image.
  • versioned_docs/version-0.18/... (both files) -- identical copies of the above, kept byte-for-byte in sync with the docs/ versions.

Tests

No automated tests here since this is a doc change, but I checked every link by hand: every kserve.github.io/website/... URL resolves with a live curl -I, and I diffed the docs/ and versioned_docs/version-0.18/ copies against each other to confirm they match.

The companion kserve/kserve PR's LMCache+Valkey sample was validated end-to-end on a GPU instance (g5.2xlarge, A10G): vLLM + LMCache successfully stored KV cache chunks to a live Valkey container and retrieved them on subsequent requests (confirmed via valkey-cli KEYS and LMCache's hit tokens logs).

One thing worth flagging: the two links to the sample files in kserve/kserve (llmisvc-lmcache-valkey.yaml and valkey.yaml) point at files that don't exist on that repo's master yet -- they're on a branch in a companion PR. This PR is meant to merge after that one lands, at which point the links resolve. If review order ends up flipped, ping me and I'll swap those to branch URLs in the meantime.

Additional context

Companion PR: kserve/kserve (adds docs/samples/llmisvc/lmcache-valkey-offloading/, docs/samples/llmisvc/precise-prefix-kv-cache-routing/valkey-shared-index/, and docs/samples/v1beta1/transformer/feast/valkey.yaml).

Currently, some of the links to kserve/kserve master won't resolve until the PR on that repository has been merged.

@netlify

netlify Bot commented Jul 15, 2026

Copy link
Copy Markdown

Deploy Preview for elastic-nobel-0aef7a ready!

Name Link
🔨 Latest commit 6c450ed
🔍 Latest deploy log https://app.netlify.com/projects/elastic-nobel-0aef7a/deploys/6a5802601cd7da0008683608
😎 Deploy Preview https://deploy-preview-703--elastic-nobel-0aef7a.netlify.app
📱 Preview on mobile
Toggle QR Code...

QR Code

Use your smartphone camera to open QR code link.

To edit notification comments on pull requests, go to your Netlify project configuration.

… Feast

Signed-off-by: Edward Liang <edward.liang@improving.com>
@edlng edlng force-pushed the valkey-kvcache-offloading-docs branch from 6f2b81a to 6c450ed Compare July 15, 2026 21:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant