orneryd / NornicDB / 27991433855

89%

Build Type

push

github

Committed by orneryd

Commit Message

perf(cypher): cache keyword scans and narrow stale index fallback

Add a bounded cache for default Cypher keyword scans so repeated query
execution avoids re-scanning identical clause-routing text. This reduces
Graphiti-style retrieval overhead where repeated prepared query shapes were
spending most CPU in keywordIndexFromDefault.

Also tighten the generic UNWIND batch match paths so partially stale property
indexes only trigger label scans for unresolved distinct values. Successful
index hits are preserved instead of replacing the whole batch lookup with a
scan-built map.

Changes:
- Cache default keyword scan results with a bounded map
- Preserve indexed match hits in relationship merge batches
- Preserve indexed match hits in multi-match-create batches
- Add regression coverage for incomplete indexed MATCH buckets

Benchmarks:
- Graphiti node similarity: ~14.3us/op -> ~6.8us/op
- Graphiti edge similarity: ~21.0us/op -> ~11.9us/op
- Graphiti chunk similarity: ~12.3us/op -> ~6.1us/op

Coverage Stats

33 of 33 new or added lines in 3 files covered. (100.0%)

11 existing lines in 6 files now uncovered.

142143 of 159807 relevant lines covered (88.95%)

1.04 hits per line