orneryd / NornicDB / 27933187127

Builds Branch Commit Type Ran Committer Via Coverage
27933187127 main fix(cypher): preserve indexed relationship batch matches Fix the generic UNWIND relationship MERGE batch path when multiple MATCH clauses share the same label/property key but read from different row fields, as in Graphiti's source/target Entity ... push orneryd github
89.01
27895411842 main perf(cypher): improve graphiti cypher query performance with hot path optimizations push orneryd github
88.99
27894591005 main fix(search): respect lazy warming for write-side indexing Keep search write events from starting BuildIndexes when a database is configured for lazy BM25/vector warming. Queued write mutations now stay parked without rescheduling flush timers unt... push orneryd github
89.09
27890936986 main fix(storage): block prefix counts during async flush NodeCountByPrefix could skip the flush lock and observe a mixed cache/engine state while async writes were in flight. That made namespace-scoped counts flaky and could report MATCH (n) lower th... push orneryd github
89.09
27887643810 main fix(search): stop background vector work after shutdown Cancel search-service background work and close cached search services during DB shutdown so HNSW maintenance, deferred rebuilds, and persist timers cannot keep running after clients disconn... push orneryd github
89.09
27860008857 main fix(search): make CPU brute-force vector search opt-in Default vector search now uses HNSW regardless of dataset size by setting the CPU brute-force threshold default to 0. Operators can opt back into exact CPU brute-force below a chosen vector c... push orneryd github
89.11
27859777643 main fix(search): keep Graphiti vector reads and writes passive Prevent live write mutations from scheduling background vector strategy transitions by default, so Graphiti-style bulk ingest does not trigger a brute-force to HNSW rebuild while requests... push orneryd github
89.16
27859488484 main fix(cypher): keep vector reads passive during Graphiti ingest Keep scalar vector.similarity cosine fast paths from triggering search warmup or falling back to exact storage scans while live vectors are already indexed. Use live node and relations... push orneryd github
89.17
27855817225 main adding additional regression tests for graphiti scenarios push orneryd github
89.17
27855696466 main fix(search,cypher): separate ONLINE readiness from live vector-query capability Keep IsReady() as full build/warmup readiness (Neo4j-like ONLINE semantics), and add CanServeVectorQueries() for the fast path to use when live indexed vectors exist b... push orneryd github
89.16
27854303697 main fix(cypher): make vector fast paths use owned search readiness Avoid query-time vector service creation while preserving deterministic indexed performance for wired database services. Changes: - Never create or replace search services from Cyphe... push orneryd github
89.16
27846551178 main fix(cypher): nested datetime fix. The root cause was valueToLiteral falling through to fmt.Sprintf("'%v'", v) for time.Time values when row.created_at was expanded from an UNWIND map. It now emits typed Cypher temporal literals: datetime('<RFC3339... push orneryd github
89.16
27845625031 main Improve performance-critical query paths without changing procedure semantics. Changes: - Keep MATCH+WITH vector cosine fast path cheap for property-free shapes, while supporting inline node property maps. - Add fast predicate handling for `<var>... push orneryd github
89.17
27840746422 main fix(cypher): index properly on node mutations push orneryd github
89.17
27831871353 main ix(cypher): execute WITH-tail relationship vector CALLs and bind rel vars" -m "Fix MATCH ... WITH r CALL db.create.setRelationshipVectorProperty(...) paths where relationship-bound CALL side effects were skipped and rel variables were not rewritte... push orneryd github
89.16