OpenSearch 主栈:专用在线检索与独立 Catalog
OpenSearch BM25 + k-NN + efficient filters · PostgreSQL control plane · S3 source of truth
OpenSearch 主栈:专用在线检索与独立 Catalog
OpenSearch serves online lexical, vector and hybrid retrieval. S3 retains original documents and parsed artifacts; a small PostgreSQL catalog controls versions, ACL and outbox state. Search API injects authorization filters before the OpenSearch query.
OpenSearch 主栈:专用在线检索与独立 Catalog
OpenSearch BM25 + k-NN + efficient filters · PostgreSQL control plane · S3 source of truth
01 · INGESTION / INDEX BUILD
02 · ONLINE QUERY / EVIDENCE
事件
解析
原件
投影
索引
请求
身份
查询构建
Hybrid
引用
INDEX QUERY
SOURCE
来源系统
DB CDC · Vendor API
PDF · Office · Email · Audio
INGEST
Connectors
Kafka / SQS / MSK
增量水位 · Replay · DLQ
WORKER
处理 Worker
Parser · OCR · ASR
Normalize · Chunk · Embed*
DURABLE
S3 Object Store
原件 · 版本 · 派生产物
Snapshot · Manifest
CATALOG
Catalog / Control
Document · Version · ACL
PostgreSQL · Outbox
INDEX
OpenSearch Index
BM25 · k-NN · Metadata
Alias · ISM · Snapshot
CALLER
用户 / Agent
UI · SDK · MCP
不直连集群
API
Search API
Auth · Rate limit · Audit
统一查询契约
SECURITY
权限注入
Tenant · ACL principals
来源 · 时间 · 市场
QUERY
Hybrid Builder
BM25 + k-NN
Filter · RRF · rerank
SERVE
OpenSearch Query
Query DSL · Search Pipeline
Efficient k-NN filtering
EVIDENCE
证据回取
Chunk / metadata → Catalog
原文 / 页码 → S3
* Embedding 是按评测决定的可选通道;OpenSearch 负责在线查询投影,不作为原始文档唯一事实来源。
REFERENCE ARCHITECTURE · LOGICAL VIEW