Accept arbitrary HMA page boundaries with CUDA restore configuration - #29
Closed
FujitsuPolycom wants to merge 1 commit into
Closed
Conversation
Page-semantic extensions accept authenticated chunk boundaries that fall inside a larger HMA page. The delta reuses only byte-identical page prefixes, so a changed boundary-intersecting page and every following page are immutable replacement data. This preserves the longest verified publication base instead of falling back to a full snapshot. Restore continues to authenticate the embedded base graph and reconstructed result before placement; invalid geometry or bytes remain a cache miss and recomputation. Canonical SparkCache CUDA restore and placement configuration remains the only generated configuration vocabulary. Deployable SparkCache source SHA-256: bc7cae86732c869ee8b2205d48ac5be6f580ee8b77a3e4ffd4c69dcd4f1bfae5. Cache namespace impact: none. CacheIdentity values, digest salts, chunk geometry, and page-delta wire schemas are unchanged. Validation: python -m pytest sparkcache -q (747 passed, 7 skipped); python -m pytest deploy -q (108 passed, 1 skipped); python -m ruff check .; source-attestation tests (2 passed).
This was referenced Aug 30, 2026
Owner
Author
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Resulting behavior
Page-semantic HMA extensions accept an authenticated cache boundary inside a larger group page. The codec compares complete opaque pages and carries a changed boundary-intersecting page, plus every following page, as immutable replacement data. Earlier byte-identical pages remain shared with the verified base snapshot.
Generated configuration continues to use only the canonical
spark_cache_cuda_restoreandspark_cache_cuda_placement_*keys supplied by the parent branch.Correctness
Restore authenticates the embedded base manifest, delta objects, layout, block counts, semantic token boundaries, and reconstructed result digest before SparkCache CUDA placement. Invalid metadata or bytes still produce a cache miss and recomputation.
The GPU-free D-16 regression exercises
SparkContextCacheConnectorwith the realManifestStore: it stores and restores a 7,168-token GLM-shaped base, publishes a 12,032-token page delta, verifies that the encoded page payload remains smaller than a full snapshot, and restores full-attention and recurrent state byte-for-byte.Compatibility
Cache namespace impact: none.
CacheIdentityvalues, digest salts, chunk geometry, and thesparkcache-hybrid-page-delta/v1andsparkcache-page-delta-manifest/v1wire schemas are unchanged.Deployable SparkCache source SHA-256:
bc7cae86732c869ee8b2205d48ac5be6f580ee8b77a3e4ffd4c69dcd4f1bfae5.This draft is stacked on #26 at commit
b95aa8ab0068dc66a6892a5c311d7e9dd4a9c55a.Distinct fail-closed condition
The error
selected page window contains vLLM's null blockoccurs while the connector selects physical HMA blocks, before snapshot or page-delta encoding. This change does not alter that fail-closed condition. A separate regression must capture the recurrent group table and its authoritative checkpoint slot before any fix chooses a non-null block; scanning for an arbitrary non-null entry would not prove state identity.Validation
python -m pytest sparkcache -q— 747 passed, 7 skippedpython -m pytest deploy -q— 108 passed, 1 skippedpython -m ruff check .— passed