Skip to content

docs: KV events + Mooncake Store integration plan#550

Draft
stmatengss wants to merge 1 commit into
mainfrom
cursor/kv-events-mooncake-plan-e3df
Draft

docs: KV events + Mooncake Store integration plan#550
stmatengss wants to merge 1 commit into
mainfrom
cursor/kv-events-mooncake-plan-e3df

Conversation

@stmatengss

Copy link
Copy Markdown

Summary

Adds an implementation plan for publishing standards-aligned KV cache events across all Mooncake Store memory tiers (GPU, host/L2, Mooncake L3), enabling cache-aware routing with Dynamo KV router and Mooncake KvIndexer (RFC #1527).

Plan location

docs/superpowers/plans/2026-06-28-kv-events-mooncake-store.md

Key points

  • Current state: TokenSpeed already publishes device-tier (GPU) events via ZMQ in Dynamo-compatible msgpack format; Mooncake L3 backup/prefetch works but emits no KV events.
  • Gaps: No medium tier tagging, FNV/SHA256 hash mismatch vs RFC #1527 XXH3-64, no host or L3 event streams.
  • Phased approach: Hash unification → RFC envelope → host/cpu events → engine L3 events → optional Mooncake master relay → Dynamo indexer validation.

References

Open in Web Open in Cursor 

Signed-off-by: Cursor Agent <cursoragent@cursor.com>

Co-authored-by: Teng Ma <stmatengss@users.noreply.github.com>
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.

2 participants