Skip to content

blockifier: preserve Native thread-local caches across sequential batches - #4

Draft
Mohiiit wants to merge 1 commit into
perf/blockifier-hash-cache-runtime-configfrom
perf/persistent-native-executor
Draft

Mohiiit wants to merge 1 commit into
perf/blockifier-hash-cache-runtime-configfrom
perf/persistent-native-executor

Conversation

@Mohiiit

@Mohiiit Mohiiit commented Sep 12, 2026

Copy link
Copy Markdown
Collaborator

Adds an opt-in stack-sized persistent execution-thread entrypoint. Sequential execution reuses only a registered worker with sufficient stack; ordinary callers and larger stack requests retain the scoped-thread fallback. Concurrent execution and Cairo Native cache implementation are unchanged.

Based on 4c5fd9e, the exact dependency of Madara continuous-nonce batching PR starkware-libs#1255. Companion Madara draft starkware-libs#1256 (madara-alliance/madara#1256) wires its existing executor loop through this entrypoint.

Validation: six standalone Rust 1.89 thread-lifecycle tests passed, including borrowed fallback work, TLS reuse, larger-stack fallback, child-thread isolation, panic propagation and replacement isolation. Non-Native helper compilation passed. Full Madara integration cargo check and four Native-enabled/disabled rejection/reversion/nonce/shutdown regressions passed in release workflow 34696401227. Seven benchmark cells passed; the largest portfolio cell OOM-killed at its unchanged 12 GiB limit after 3,999 of 5,000 invokes. Keep draft: memory retention attribution/profiling and a successful full PM150 run remain unresolved. See the companion draft for exact numbers and limitations. No Cairo Native implementation changes.

@Mohiiit Mohiiit changed the title perf(blockifier): preserve Native thread-local caches across sequential batches blockifier: preserve Native thread-local caches across sequential batches Sep 12, 2026
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