目录
1、背景
我们知道当我们使用 terms聚合时，当修改默认顺序为_count asc时，统计的结果是不准备的，而且官方也不推荐我们这样做，而是推荐使用rare terms聚合。rare terms是一个稀少的term聚合，可以一定程度的解决升序问题。
2、需求
统计province字段中包含上和湖的term数据，并且最多只能出现2次。获取到聚合后的结果。
3、前置准备
3.1 准备mapping
PUT /index_person { "settings": { "number_of_shards": 1 }, "mappings": { "properties": { "id": { "type": "long" }, "name": { "type": "keyword" }, "province": { "type": "keyword" }, "sex": { "type": "keyword" }, "age": { "type": "integer" }, "pipeline_province_sex":{ "type": "keyword" }, "address": { "type": "text", "analyzer": "ik_max_word", "fields": { "keyword": { "type": "keyword", "ignore_above": 256 } } } } } } 3.2 准备数据
PUT /_bulk {"create":{"_index":"index_person","_id":1}} {"id":1,"name":"张三","sex":"男","age":20,"province":"湖北","address":"湖北省黄冈市罗田县匡河镇"} {"create":{"_index":"index_person","_id":2}} {"id":2,"name":"李四","sex":"男","age":19,"province":"江苏","address":"江苏省南京市"} {"create":{"_index":"index_person","_id":3}} {"id":3,"name":"王武","sex":"女","age":25,"province":"湖北","address":"湖北省武汉市江汉区"} {"create":{"_index":"index_person","_id":4}} {"id":4,"name":"赵六","sex":"女","age":30,"province":"北京","address":"北京市东城区"} {"create":{"_index":"index_person","_id":5}} {"id":5,"name":"钱七","sex":"女","age":16,"province":"北京","address":"北京市西城区"} {"create":{"_index":"index_person","_id":6}} {"id":6,"name":"王八","sex":"女","age":45,"province":"北京","address":"北京市朝阳区"} {"create":{"_index":"index_person","_id":7}} {"id":7,"name":"九哥","sex":"男","age":25,"province":"上海市","address":"上海市嘉定区"} 4、实现需求
4.1 dsl
GET /index_person/_search { "size": 0, "aggs": { "agg_province": { "rare_terms": { "field": "province", "max_doc_count": 2, "precision": 0.01, "include": "(.*上.*|.*湖.*|.*江.*)", "exclude": ["江苏"], "missing": "default省" } } } } 4.2 java代码
@Test @DisplayName("稀少的term聚合，类似按照 _count asc 排序的terms聚合，但是terms聚合中按照_count asc的结果是不准的，需要使用 rare terms 聚合") public void agg01() throws IOException { SearchRequest searchRequest = new SearchRequest.Builder() .size(0) .index("index_person") .aggregations("agg_province", agg -> agg.rareTerms(rare -> // 稀有词 的字段 rare.field("province") // 该稀有词最多可以出现在几个文档中，最大值为100，如果要调整，需要修改search.max_buckets参数的值(尝试修改这个值，不生效) // 在该例子中，只要是出现的次数<=2的聚合都会返回 .maxDocCount(2L) // 内部布谷鸟过滤器的精度，精度越小越准，但是相应的消耗内存也越多，最小值为 0.00001，默认值为 0.01 .precision(0.01) // 应该包含在聚合的term, 当是单个字段是，可以写正则表达式 .include(include -> include.regexp("(.*上.*|.*湖.*|.*江.*)")) // 排出在聚合中的term，当是集合时，需要写准确的值 .exclude(exclude -> exclude.terms(Collections.singletonList("江苏"))) // 当文档中缺失province字段时，给默认值 .missing("default省") ) ) .build(); System.out.println(searchRequest); SearchResponse<Object> response = client.search(searchRequest, Object.class); System.out.println(response); } 一些注意事项都在注释中。
4.3 运行结果
5、max_doc_count 和 search.max_buckets
6、注意事项
rare terms统计返回的数据没有大小限制，而且受max_doc_count参数的限制，比如：如果复合 max_doc_count 的分组有60个，那么这60个分组会直接返回。max_doc_count的值最大为100，貌似不能修改。- 如果一台节点聚合收集的结果过多，那么很容易超过
search.max_buckets的值，此时就需要修改这个值。
# 临时修改 PUT /_cluster/settings {"transient": {"search.max_buckets": 65536}} # 永久修改 PUT /_cluster/settings {"persistent": {"search.max_buckets": 65536}} 
