第五阶段 50 · geo 地理查询（geo_point / geo_distance）

50 · geo 地理查询geo_point / geo_distance阶段第五阶段 / 进阶可选业务涉及地理位置时ESgeo_point/geo_shapegeo_distance/geo_bounding_box| PostgreSQLPostGISgeography/ST_DWithin1. 概念存经纬度并按「距离/范围」检索就用地理类型类型用途geo_point一个点经纬度最常用geo_shape线/面/多边形等复杂形状常见查询geo_distance某点 N 公里内、geo_bounding_box矩形框内、geo_distance聚合/排序按距离排序、分环统计。2. PostgreSQL 对照-- PostGIS5km 内的门店SELECT*FROMstoresWHEREST_DWithin(geog,ST_MakePoint(:lon,:lat)::geography,5000);-- 按距离排序SELECT*,geog-ST_MakePoint(:lon,:lat)::geographyASdistFROMstoresORDERBYdistLIMIT10;ES 的geo_distance≈ST_DWithin按距离排序 ≈-排序。3. ES DSL3.1 mapping 写入PUT stores_idx { mappings: { properties: { name: { type: keyword }, location: { type: geo_point } } } } POST stores_idx/_doc { name: Store-1, location: { lat: 39.9042, lon: 116.4074 } }3.2 geo_distanceN 公里内 按距离排序GET stores_idx/_search { query: { bool: { filter: { geo_distance: { distance: 5km, location: { lat: 39.9, lon: 116.4 } } } } }, sort: [ { _geo_distance: { location: { lat: 39.9, lon: 116.4 }, order: asc, unit: km } } ] }3.3 geo_bounding_box矩形框内query: { geo_bounding_box: { location: { top_left: { lat: 40.1, lon: 116.2 }, bottom_right: { lat: 39.7, lon: 116.6 } } } }4. Spring Boot 实现ComponentpublicclassDoc50GeoQuery{AutowiredprivateElasticsearchClientelasticsearchClient;/** 某点 radiusKm 公里内的门店按距离升序 */publicListMapString,Objectnearby(StringindexName,doublelat,doublelon,doubleradiusKm)throwsIOException{SearchResponseMaprespelasticsearchClient.search(s-s.index(indexName).query(q-q.bool(b-b.filter(f-f.geoDistance(g-g.field(location).distance(radiusKmkm).location(l-l.latlon(ll-ll.lat(lat).lon(lon))))))).sort(so-so.geoDistance(gd-gd.field(location).location(l-l.latlon(ll-ll.lat(lat).lon(lon))).order(SortOrder.Asc).unit(DistanceUnit.Kilometers))),Map.class);returnresp.hits().hits().stream().map(Hit::source).filter(Objects::nonNull).collect(Collectors.toList());}}importco.elastic.clients.elasticsearch._types.SortOrder、...DistanceUnit。每个 hit 的距离在hit.sort()对应_geo_distance排序值里可取出。5. 坑与最佳实践经纬度顺序易错GeoJSON 数组是[lon, lat]对象是{lat, lon}——别写反。geo_distance放filter不需要打分放 filter 更快、可缓存。距离排序有成本每命中文档都算距离大结果集配合先用geo_distance过滤缩小范围。精度 vs 性能geo_distance有arc精确/plane快、近距离够用两种算法。复杂形状用geo_shape点用geo_point即可多边形/围栏才上geo_shape。