Repository navigation
Fix cache isolation and admission; optimize CPU hot paths - #7
Merged
Merged
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem and changes
Rejected or replayed file-store reservations could evict valid cached data, and global idempotency lookup could return another tenant's cached mutation result. Admission now validates before reclamation, batches reject duplicate request IDs before mutation, and replay state is scoped by authenticated tenant and trust domain. A TCP client also stops accepting calls after an incomplete exchange so a late reply cannot answer a later request.
CPU hot-path changes remove per-read LRU disk syncs and two framing copies, move parsing/encoding outside the store mutex, borrow checksum inputs, and index eviction/reload and shard-directory bookkeeping. Directory replay validates tenant/size and rejects older member incarnations. Cold refill charges manifest plus payload to one budget, while simulated DPU writes share a pending/committed byte cap and move payload ownership on completion.
GCK1/GCKR and CacheIdentity encodings are unchanged. No dependencies or hardware integration are added. Documentation makes the remaining production limitations explicit, including reader pins, response allocation, unwired TCP security hooks, distributed fencing, and hidden-slab recovery. Read recency is advisory across restart; committed-block persistence is unchanged.
The cross-process test helper uses Cargo's target/profile-specific binary path and the existing Cargo target-runner environment variable. The helper reuses QEMU CI’s executable and ARM64 library-path arguments so the spawned helper executes correctly; native test execution remains unchanged.
Validation
cargo test --workspace --all-targets.cargo test --workspace --all-targets --all-features.cargo fmt --all -- --check,cargo clippy --workspace --all-targets -- -D warnings, andgit diff --checkpass.cargo run --release -p gpucache-transport --example cpu_auditfor reproducible synthetic file/directory microbenchmarks. These are not inference, NIC, or DPU measurements.RDMA/GPU/BF3 remain
NOT_RUN_HARDWARE; cost remainsTCO_UNKNOWN.