Skip to content

perf(knowledge): benchmark and harden reranked knowledge search latency #595

@ohdearquant

Description

@ohdearquant

Body:

  • Problem: knowledge.search(rerank=true, limit=8) timed out at 120 seconds during survey synthesis and prior memory also records slow first-search/warm-load behavior, degrading research grounding throughput.
  • Proposed fix: add a reproducible benchmark for reranked search latency, separate cold-start from steady-state timings, and identify whether timeout risk is Vamana warm load, rerank scoring, atom slug resolution, or MCP batching.
  • Acceptance criteria: benchmark script records cold and warm p50/p95 for reranked search; timeout reproduction is either fixed or documented with a guardrail; live atom-slug resolution has a test or diagnostic; results are saved in a perf artifact.
  • Affected crates/ADRs: crates/khive-pack-knowledge/src/knowledge/mod.rs:1106, crates/khive-pack-knowledge/src/knowledge/mod.rs:1893, /Users/lion/khive-work/shows/khive-improvement-survey/feature-frontier/synthesis.md:90, /Users/lion/khive-work/shows/khive-improvement-survey/feature-frontier/synthesis.md:93, memory note 9e1b82f9, memory note 72f9fdda.

Source: khive-improvement-survey show — proposed backlog (propose-only synthesis). Dedup: NEW.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingcrate/knowledgekhive-pack-knowledge cratecrate/runtimekhive-runtime cratetype/perfPerformance improvement

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions