Repository navigation
Lazy KG loading with LRU eviction; O(shards) startup #135
Description
load_all_knowledge_graphs loads every KG eagerly and scans all shards once per KG (O(KGs x shards)), with JSON WAL replay. With one KG per user, startup time and RSS grow with the total data of all users. Group shards by KG once, load on first access, evict idle KGs, and load in parallel.
Roadmap checkpoint — 2026-10-03
R02/R03 remain prerequisites for D08 (#187). Shard grouping already exists, but src/naming.rs:57 scans known KG names per shard; measure and replace that ownership lookup rather than only parallelizing eager loading. Implement lazy activation, lifetime pinning for active requests/subscriptions, bounded safe eviction and metadata-only startup. Avoid a per-request LRU mutex; pass warm-query/delta and cold-start/RSS gates.
Activity
perf(storage): index shard ownership so startup is linear in shards
perf(storage): index shard ownership so startup is linear in shards (#…
jsam commentedon Oct 4, 2026
Moved to the Engine production readiness milestone, the current focus (tracker #199). EN-13: fast restart and measured restart time (lazy loading, O(shards) startup).
Roadmap per the neurosymbolic decision layer plan: engine production readiness first, then gateway neural integration with Jev, CLM and Laya (#266).
feat(storage): load knowledge graphs on first use so restart reads on…
jsam commentedon Oct 5, 2026
Done in #357 (merged): lazy graph activation and O(shards) startup.