ElasticSearch进阶:各种ES查询在Java中的实现
注:本文摘自:https://mp.weixin.qq.com/s/7vEy-vN8JV3o6sAh6HFohA
本文基于elasticsearch 7.13.2版本,es从7.0以后,发生了很大的更新。7.3以后,已经不推荐使用TransportClient这个client,取而代之的是Java High Level REST Client。
测试使用的数据示例
首先是,Mysql中的部分测试数据:
Mysql中的一行数据在ES中以一个文档形式存在:
{
"_index" : "person",
"_type" : "_doc",
"_id" : "4",
"_score" : 1.0,
"_source" : {
"address" : "峨眉山",
"modifyTime" : "2021-06-29 19:46:25",
"createTime" : "2021-05-14 11:37:07",
"sect" : "峨嵋派",
"sex" : "男",
"skill" : "降龙十八掌",
"name" : "宋青书",
"id" : 4,
"power" : 50,
"age" : 21
}
}
简单梳理了一下ES JavaAPI的相关体系,感兴趣的可以自己研读一下源码。
接下来,我们用十几个实例,迅速上手ES的查询操作,每个示例将提供SQL语句、ES语句和Java代码。
1 词条查询
所谓词条查询,也就是ES不会对查询条件进行分词处理,只有当词条和查询字符串完全匹配时,才会被查询到。
1.1 等值查询-term
等值查询,即筛选出一个字段等于特定值的所有记录。
SQL:
select * from person where name = '张无忌';
而使用ES查询语句却很不一样(注意查询字段带上keyword):
GET /person/_search
{
"query": {
"term": {
"name.keyword": {
"value": "张无忌",
"boost": 1.0
}
}
}
}
ElasticSearch 5.0以后,string类型有重大变更,移除了string类型,string字段被拆分成两种新的数据类型: text用于全文搜索的,而keyword用于关键词搜索。
查询结果:
{
"took" : 0,
"timed_out" : false,
"_shards" : { // 分片信息
"total" : 1, // 总计分片数
"successful" : 1, // 查询成功的分片数
"skipped" : 0, // 跳过查询的分片数
"failed" : 0 // 查询失败的分片数
},
"hits" : { // 命中结果
"total" : {
"value" : 1, // 数量
"relation" : "eq" // 关系:等于
},
"max_score" : 2.8526313, // 最高分数
"hits" : [
{
"_index" : "person", // 索引
"_type" : "_doc", // 类型
"_id" : "1",
"_score" : 2.8526313,
"_source" : {
"address" : "光明顶",
"modifyTime" : "2021-06-29 16:48:56",
"createTime" : "2021-05-14 16:50:33",
"sect" : "明教",
"sex" : "男",
"skill" : "九阳神功",
"name" : "张无忌",
"id" : 1,
"power" : 99,
"age" : 18
}
}
]
}
}
Java中构造ES请求的方式:(后续例子中只保留SearchSourceBuilder的构建语句)
/**
* term精确查询
*
* @throws IOException
*/
@Autowired
private RestHighLevelClient client;
@Test
public void queryTerm() throws IOException {
// 根据索引创建查询请求
SearchRequest searchRequest = new SearchRequest("person");
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.termQuery("name.keyword", "张无忌"));
System.out.println("searchSourceBuilder=====================" + searchSourceBuilder);
searchRequest.source(searchSourceBuilder);
SearchResponse response = client.search(searchRequest, RequestOptions.DEFAULT);
System.out.println(JSONObject.toJSON(response));
}
仔细观察查询结果,会发现ES查询结果中会带有_score
这一项,ES会根据结果匹配程度进行评分。打分是会耗费性能的,如果确认自己的查询不需要评分,就设置查询语句关闭评分:
GET /person/_search
{
"query": {
"constant_score": {
"filter": {
"term": {
"sect.keyword": {
"value": "张无忌",
"boost": 1.0
}
}
},
"boost": 1.0
}
}
}
Java构建查询语句:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 这样构造的查询条件,将不进行score计算,从而提高查询效率
searchSourceBuilder.query(QueryBuilders.constantScoreQuery(QueryBuilders.termQuery("sect.keyword", "明教")));
1.2 多值查询-terms
多条件查询类似Mysql里的IN查询,例如:
select * from persons where sect in('明教','武当派');
ES查询语句:
GET /person/_search
{
"query": {
"terms": {
"sect.keyword": [
"明教",
"武当派"
],
"boost": 1.0
}
}
}
Java实现:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.termsQuery("sect.keyword", Arrays.asList("明教", "武当派")));
}
1.3 范围查询-range
范围查询,即查询某字段在特定区间的记录。
SQL:
select * from pesons where age between 18 and 22;
ES查询语句:
GET /person/_search
{
"query": {
"range": {
"age": {
"from": 10,
"to": 20,
"include_lower": true,
"include_upper": true,
"boost": 1.0
}
}
}
}
Java构建查询条件:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.rangeQuery("age").gte(10).lte(30));
}
1.4 前缀查询-prefix
前缀查询类似于SQL中的模糊查询。
SQL:
select * from persons where sect like '武当%';
ES查询语句:
{
"query": {
"prefix": {
"sect.keyword": {
"value": "武当",
"boost": 1.0
}
}
}
}
Java构建查询条件:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.prefixQuery("sect.keyword","武当"));
1.5 通配符查询-wildcard
通配符查询,与前缀查询类似,都属于模糊查询的范畴,但通配符显然功能更强。
SQL:
select * from persons where name like '张%忌';
ES查询语句:
{
"query": {
"wildcard": {
"sect.keyword": {
"wildcard": "张*忌",
"boost": 1.0
}
}
}
}
Java构建查询条件:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.wildcardQuery("sect.keyword","张*忌"));
2 复合查询
前面的例子都是单个条件查询,在实际应用中,我们很有可能会过滤多个值或字段。先看一个简单的例子:
select * from persons where sex = '女' and sect = '明教';
这样的多条件等值查询,就要借用到组合过滤器了,其查询语句是:
{
"query": {
"bool": {
"must": [
{
"term": {
"sex": {
"value": "女",
"boost": 1.0
}
}
},
{
"term": {
"sect.keywords": {
"value": "明教",
"boost": 1.0
}
}
}
],
"adjust_pure_negative": true,
"boost": 1.0
}
}
}
Java构造查询语句:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.boolQuery()
.must(QueryBuilders.termQuery("sex", "女"))
.must(QueryBuilders.termQuery("sect.keyword", "明教"))
);
2.1 布尔查询
布尔过滤器(bool filter)属于复合过滤器(compound filter)的一种 ,可以接受多个其他过滤器作为参数,并将这些过滤器结合成各式各样的布尔(逻辑)组合。
bool 过滤器下可以有4种子条件,可以任选其中任意一个或多个。filter是比较特殊的,这里先不说。
{
"bool" : {
"must" : [],
"should" : [],
"must_not" : [],
}
}
- must:所有的语句都必须匹配,与 ‘=’ 等价。
- must_not:所有的语句都不能匹配,与 ‘!=’ 或 not in 等价。
- should:至少有n个语句要匹配,n由参数控制。
精度控制:
所有 must 语句必须匹配,所有 must_not
语句都必须不匹配,但有多少 should 语句应该匹配呢?默认情况下,没有 should 语句是必须匹配的,只有一个例外:那就是当没有 must 语句的时候,至少有一个 should 语句必须匹配。
我们可以通过 minimum_should_match
参数控制需要匹配的 should 语句的数量,它既可以是一个绝对的数字,又可以是个百分比:
GET /person/_search
{
"query": {
"bool": {
"must": [
{
"term": {
"sex": {
"value": "女",
"boost": 1.0
}
}
}
],
"should": [
{
"term": {
"address.keyword": {
"value": "峨眉山",
"boost": 1.0
}
}
},
{
"term": {
"sect.keyword": {
"value": "明教",
"boost": 1.0
}
}
}
],
"adjust_pure_negative": true,
"minimum_should_match": "1",
"boost": 1.0
}
}
}
Java构建查询语句:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.boolQuery()
.must(QueryBuilders.termQuery("sex", "女"))
.should(QueryBuilders.termQuery("address.word", "峨眉山"))
.should(QueryBuilders.termQuery("sect.keyword", "明教"))
.minimumShouldMatch(1)
);
最后,看一个复杂些的例子,将bool的各子句联合使用:
select
*
from
persons
where
sex = '女'
and
age between 30 and 40
and
sect != '明教'
and
(address = '峨眉山' OR skill = '暗器')
用 Elasticsearch 来表示上面的 SQL 例子:
GET /person/_search
{
"query": {
"bool": {
"must": [
{
"term": {
"sex": {
"value": "女",
"boost": 1.0
}
}
},
{
"range": {
"age": {
"from": 30,
"to": 40,
"include_lower": true,
"include_upper": true,
"boost": 1.0
}
}
}
],
"must_not": [
{
"term": {
"sect.keyword": {
"value": "明教",
"boost": 1.0
}
}
}
],
"should": [
{
"term": {
"address.keyword": {
"value": "峨眉山",
"boost": 1.0
}
}
},
{
"term": {
"skill.keyword": {
"value": "暗器",
"boost": 1.0
}
}
}
],
"adjust_pure_negative": true,
"minimum_should_match": "1",
"boost": 1.0
}
}
}
用Java构建这个查询条件:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
BoolQueryBuilder boolQueryBuilder = QueryBuilders.boolQuery()
.must(QueryBuilders.termQuery("sex", "女"))
.must(QueryBuilders.rangeQuery("age").gte(30).lte(40))
.mustNot(QueryBuilders.termQuery("sect.keyword", "明教"))
.should(QueryBuilders.termQuery("address.keyword", "峨眉山"))
.should(QueryBuilders.rangeQuery("power.keyword").gte(50).lte(80))
.minimumShouldMatch(1); // 设置should至少需要满足几个条件
// 将BoolQueryBuilder构建到SearchSourceBuilder中
searchSourceBuilder.query(boolQueryBuilder);
2.2 Filter查询
query和filter的区别:query查询的时候,会先比较查询条件,然后计算分值,最后返回文档结果;而filter是先判断是否满足查询条件,如果不满足会缓存查询结果(记录该文档不满足结果),满足的话,就直接缓存结果,filter不会对结果进行评分,能够提高查询效率。
filter的使用方式比较多样,下面用几个例子演示一下。
方式一,单独使用:
{ "query": { "bool": { "filter": [ { "term": { "sex": { "value": "男", "boost": 1.0 } } } ], "adjust_pure_negative": true, "boost": 1.0 } } }
单独使用时,filter与must基本一样,不同的是filter不计算评分,效率更高。
Java构建查询语句:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.boolQuery()
.filter(QueryBuilders.termQuery("sex", "男"))
);
方式二,和must、must_not同级,相当于子查询:
select * from (select * from persons where sect = '明教')) a where sex = '女';
ES查询语句:
{
"query": {
"bool": {
"must": [
{
"term": {
"sect.keyword": {
"value": "明教",
"boost": 1.0
}
}
}
],
"filter": [
{
"term": {
"sex": {
"value": "女",
"boost": 1.0
}
}
}
],
"adjust_pure_negative": true,
"boost": 1.0
}
}
}
Java:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.boolQuery()
.must(QueryBuilders.termQuery("sect.keyword", "明教"))
.filter(QueryBuilders.termQuery("sex", "女"))
);
方式三,将must、must_not置于filter下,这种方式是最常用的:
{
"query": {
"bool": {
"filter": [
{
"bool": {
"must": [
{
"term": {
"sect.keyword": {
"value": "明教",
"boost": 1.0
}
}
},
{
"range": {
"age": {
"from": 20,
"to": 35,
"include_lower": true,
"include_upper": true,
"boost": 1.0
}
}
}
],
"must_not": [
{
"term": {
"sex.keyword": {
"value": "女",
"boost": 1.0
}
}
}
],
"adjust_pure_negative": true,
"boost": 1.0
}
}
],
"adjust_pure_negative": true,
"boost": 1.0
}
}
}
Java:
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 构建查询语句
searchSourceBuilder.query(QueryBuilders.boolQuery()
.filter(QueryBuilders.boolQuery()
.must(QueryBuilders.termQuery("sect.keyword", "明教"))
.must(QueryBuilders.rangeQuery("age").gte(20).lte(35))
.mustNot(QueryBuilders.termQuery("sex.keyword", "女")))
);
3 聚合查询
接下来,我们将用一些案例演示ES聚合查询。
注意:在使用聚合查询的时候,如果查询字段为text类型,则可能出现报错:"illegal_argument_exception"
可以使用如下方法修改其默认配置(fieldata默认为false):
curl -H "Content-Type:application/json" -XPUT 'localhost:9200/article/_mapping' -d'
> {
> "properties": {
> "author": {
> "type": "text",
> "fielddata": true
> }
> }
> }'
3.1 最值、平均值、求和
案例:查询最大年龄、最小年龄、平均年龄。
SQL:
select max(age) from persons;
ES:
GET /person/_search
{
"aggregations": {
"max_age": {
"max": {
"field": "age"
}
}
}
}
Java:
@Autowired
private RestHighLevelClient client;
@Test
public void maxQueryTest() throws IOException {
// 聚合查询条件
AggregationBuilder aggBuilder = AggregationBuilders.max("max_age").field("age");
SearchRequest searchRequest = new SearchRequest("person");
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 将聚合查询条件构建到SearchSourceBuilder中
searchSourceBuilder.aggregation(aggBuilder);
System.out.println("searchSourceBuilder----->" + searchSourceBuilder);
searchRequest.source( );
// 执行查询,获取SearchResponse
SearchResponse response = client.search(searchRequest, RequestOptions.DEFAULT);
System.out.println(JSONObject.toJSON(response));
}
使用聚合查询,结果中默认只会返回10条文档数据(当然我们关心的是聚合的结果,而非文档)。返回多少条数据可以自主控制:
GET /person/_search { "size": 20, "aggregations": { "max_age": { "max": { "field": "age" } } } }
而Java中只需增加下面一条语句即可:
searchSourceBuilder.size(20);
与max类似,其他统计查询也很简单:
AggregationBuilder minBuilder = AggregationBuilders.min("min_age").field("age");
AggregationBuilder avgBuilder = AggregationBuilders.avg("min_age").field("age");
AggregationBuilder sumBuilder = AggregationBuilders.sum("min_age").field("age");
AggregationBuilder countBuilder = AggregationBuilders.count("min_age").field("age");
3.2 去重查询
案例:查询一共有多少个门派。
SQL:
select count(distinct sect) from persons;
ES:
{
"aggregations": {
"sect_count": {
"cardinality": {
"field": "sect.keyword"
}
}
}
}
Java:
@Test
public void cardinalityQueryTest() throws IOException {
// 创建某个索引的request
SearchRequest searchRequest = new SearchRequest("person");
// 查询条件
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 聚合查询
AggregationBuilder aggBuilder = AggregationBuilders.cardinality("sect_count").field("sect.keyword");
searchSourceBuilder.size(0);
// 将聚合查询构建到查询条件中
searchSourceBuilder.aggregation(aggBuilder);
System.out.println("searchSourceBuilder----->" + searchSourceBuilder);
searchRequest.source(searchSourceBuilder);
// 执行查询,获取结果
SearchResponse response = client.search(searchRequest, RequestOptions.DEFAULT);
System.out.println(JSONObject.toJSON(response));
}
3.3 分组聚合
3.3.1 单条件分组
案例:查询每个门派的人数
SQL:
select sect,count(id) from mytest.persons group by sect;
ES:
{
"size": 0,
"aggregations": {
"sect_count": {
"terms": {
"field": "sect.keyword",
"size": 10,
"min_doc_count": 1,
"shard_min_doc_count": 0,
"show_term_doc_count_error": false,
"order": [
{
"_count": "desc"
},
{
"_key": "asc"
}
]
}
}
}
}
Java:
SearchRequest searchRequest = new SearchRequest("person");
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
searchSourceBuilder.size(0);
// 按sect分组
AggregationBuilder aggBuilder = AggregationBuilders.terms("sect_count").field("sect.keyword");
searchSourceBuilder.aggregation(aggBuilder);
3.3.2 多条件分组
案例:查询每个门派各有多少个男性和女性
SQL:
select sect,sex,count(id) from mytest.persons group by sect,sex;
ES:
{
"aggregations": {
"sect_count": {
"terms": {
"field": "sect.keyword",
"size": 10
},
"aggregations": {
"sex_count": {
"terms": {
"field": "sex.keyword",
"size": 10
}
}
}
}
}
}
3.4 过滤聚合
前面所有聚合的例子请求都省略了 query ,整个请求只不过是一个聚合。这意味着我们对全部数据进行了聚合,但现实应用中,我们常常对特定范围的数据进行聚合,例如下例。
案例:查询明教中的最大年龄。这涉及到聚合与条件查询一起使用。
SQL:
select max(age) from mytest.persons where sect = '明教';
ES:
GET /person/_search
{
"query": {
"term": {
"sect.keyword": {
"value": "明教",
"boost": 1.0
}
}
},
"aggregations": {
"max_age": {
"max": {
"field": "age"
}
}
}
}
Java:
SearchRequest searchRequest = new SearchRequest("person");
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
// 聚合查询条件
AggregationBuilder maxBuilder = AggregationBuilders.max("max_age").field("age");
// 等值查询
searchSourceBuilder.query(QueryBuilders.termQuery("sect.keyword", "明教"));
searchSourceBuilder.aggregation(maxBuilder);
另外还有一些更复杂的查询例子。
案例:查询0-20,21-40,41-60,61以上的各有多少人。
SQL:
select
sum(case when age<=20 then 1 else 0 end) ageGroup1,
sum(case when age >20 and age <=40 then 1 else 0 end) ageGroup2,
sum(case when age >40 and age <=60 then 1 else 0 end) ageGroup3,
sum(case when age >60 and age <=200 then 1 else 0 end) ageGroup4
from
mytest.persons;
ES:
{
"size": 0,
"aggregations": {
"age_avg": {
"range": {
"field": "age",
"ranges": [
{
"from": 0.0,
"to": 20.0
},
{
"from": 21.0,
"to": 40.0
},
{
"from": 41.0,
"to": 60.0
},
{
"from": 61.0,
"to": 200.0
}
],
"keyed": false
}
}
}
}
查询结果:
"aggregations" : {
"age_avg" : {
"buckets" : [
{
"key" : "0.0-20.0",
"from" : 0.0,
"to" : 20.0,
"doc_count" : 3
},
{
"key" : "21.0-40.0",
"from" : 21.0,
"to" : 40.0,
"doc_count" : 13
},
{
"key" : "41.0-60.0",
"from" : 41.0,
"to" : 60.0,
"doc_count" : 4
},
{
"key" : "61.0-200.0",
"from" : 61.0,
"to" : 200.0,
"doc_count" : 1
}
]
}
}