6 Commits

Include renames

Author	SHA1	Message	Date
Yuanqing ZHAOandYuanqing Zhao	fa19ac0a75	perf(vectordb): coalesce auto cuVS rebuilds during bulk ingest (#3277) ... * perf(vectordb): coalesce auto cuVS rebuilds during bulk ingest Add an opt-in bulk-ingest maintenance scope that coalesces Auto cuVS background rebuilds across multiple write batches. - defer derived GPU maintenance until the outermost bulk scope exits while keeping native writes and persistence visible per call - harden the background worker against debounce, generation, shutdown, and stale-candidate races - preserve suspension across index replacement and retire replaced workers - wait for the final Auto GPU snapshot before vectordb_perf records search QPS - document that the scope is non-transactional and only schedules readiness on exit Auto cuVS and background rebuild remain disabled by default. Native CPU and remote backends use no-op hooks, so their existing behavior and dtype are unchanged. * fix(vectordb): reject stale index replacements * fix(vectordb): harden bulk rebuild lifecycle --------- Co-authored-by: Yuanqing Zhao <2604121+yuanqingz@users.noreply.github.com>	2026-07-16 18:59:16 +08:00
Yuanqing ZHAOandYuanqing Zhao	546da35cc8	fix(benchmark): load WIKI-Dir path mapping (#3279) ... Co-authored-by: Yuanqing Zhao <2604121+yuanqingz@users.noreply.github.com>	2026-07-16 11:11:15 +08:00
Yuanqing ZHAOandYuanqing Zhao	0080f94bdc	perf(vectordb): batch benchmark upserts (#3264) ... Co-authored-by: Yuanqing Zhao <2604121+yuanqingz@users.noreply.github.com>	2026-07-15 19:49:06 +08:00
Yuanqing ZHAOandYuanqing Zhao	7e6a0515f9	perf(cuvs): optimize filters, rebuilds, concurrency, and memory (#3092) ... * perf(cuvs): fast-path cached native filter routes * perf(cuvs): parallelize auto filter preflight * perf(cuvs): add search route telemetry * test(cuvs): use a valid telemetry vector dimension * perf(cuvs): reuse native filter preflight results * perf(cuvs): allow concurrent snapshot searches * perf(cuvs): coalesce optional background rebuilds * perf(cuvs): coordinate per-GPU build admission * perf(cuvs): add opt-in float16 search * build(cuvs): support vector benchmark harnesses * perf(cuvs): bound concurrent GPU searches * perf(cuvs): avoid partial background rebuilds * fix(cuvs): address rebuild and telemetry review feedback * fix(cuvs): defer rebuild until index initialization --------- Co-authored-by: Yuanqing Zhao <2604121+yuanqingz@users.noreply.github.com>	2026-07-10 17:22:34 +08:00
Yuanqing ZHAOandYuanqing Zhao	d61d802fd0	fix(benchmark): enforce configured search concurrency (#3091) ... Co-authored-by: Yuanqing Zhao <2604121+yuanqingz@users.noreply.github.com>	2026-07-09 11:06:58 +08:00
Yuanqing ZHAOandYuanqing Zhao	39c778c953	feat: add cuVS vector search backend (#2974) ... * feat: add cuVS vector search backend * docs: add agent memory benchmark strategy * bench: add cuVS index performance harness * bench: add public ANN dataset tuning * docs: record preliminary cuVS index results * docs: clarify warm index latency * docs: order cuVS before qdrant * bench: aggregate independent index runs * bench: order aggregate variants consistently * docs: add repeatable index scaling results * bench: add collection lifecycle benchmark * docs: add collection lifecycle results * perf: cache prepared cuvs filters * docs: report prepared filter cache results * bench: add async vector concurrency benchmark * bench: aggregate service concurrency runs * docs: add async concurrency results * docs: clarify cuVS dtype behavior * feat: add memory-aware cuVS auto mode * feat: reuse native filters for cuVS search * docs: publish cuVS integration plan as Markdown * fix: route selective filters before cuVS rebuild * docs: record selective-first routing results --------- Co-authored-by: Yuanqing Zhao <2604121+yuanqingz@users.noreply.github.com>	2026-07-07 12:21:10 +08:00