以球员信息为例,player索引的player type包含5个字段,姓名,年龄,薪水,球队,场上位置。
index的mapping为:
- "mappings": {
- "quote": {
- "properties": {
- "adj_close": {
- "type": "long"
- },
- "open": {
- "type": "long"
- },
- "symbol": {
- "index": "not_analyzed",
- "type": "string"
- },
- "volume": {
- "type": "long"
- },
- "high": {
- "type": "long"
- },
- "low": {
- "type": "long"
- },
- "date": {
- "format": "strict_date_optional_time||epoch_millis",
- "type": "date"
- },
- "close": {
- "type": "long"
- }
- },
- "_all": {
- "enabled": false
- }
- }
- }
索引中的全部数据:
name | age | salary | team | position |
james | 33 | 3000 | cav | sf |
irving | 25 | 2000 | cav | pg |
curry | 29 | 1000 | war | pg |
thompson | 26 | 2000 | war | sg |
green | 26 | 2000 | war | pf |
garnett | 40 | 1000 | tim | pf |
towns | 21 | 500 | tim | c |
lavin | 21 | 300 | tim | sg |
wigins | 20 | 500 | tim | sf |
首先,初始化Builder:
- SearchRequestBuilder sbuilder = client.prepareSearch("player").setTypes("player");
接下来举例说明各种聚合操作的实现方法,因为在es的api中,多字段上的聚合操作需要用到子聚合(subAggregation),初学者可能找不到方法(网上资料比较少,笔者在这个问题上折腾了两天,最后度了源码才彻底搞清楚T_T),后边会特意说明多字段聚合的实现方法。另外,聚合后的排序也会单独说明。
1. group by/count
例如要计算每个球队的球员数,如果使用SQL语句,应表达如下:
- select team, count(*) as player_count from player group by team;
ES的java api:
- TermsBuilder teamAgg= AggregationBuilders.terms("player_count ").field("team");
- sbuilder.addAggregation(teamAgg);
- SearchResponse response = sbuilder.execute().actionGet();
2.group by多个field
例如要计算每个球队每个位置的球员数,如果使用SQL语句,应表达如下:
- select team, position, count(*) as pos_count from player group by team, position;
ES的java api:
- TermsBuilder teamAgg= AggregationBuilders.terms("player_count ").field("team");
- TermsBuilder posAgg= AggregationBuilders.terms("pos_count").field("position");
- sbuilder.addAggregation(teamAgg.subAggregation(posAgg));
- SearchResponse response = sbuilder.execute().actionGet();
3.max/min/sum/avg
例如要计算每个球队年龄最大/最小/总/平均的球员年龄,如果使用SQL语句,应表达如下:
- select team, max(age) as max_age from player group by team;
ES的java api:
- TermsBuilder teamAgg= AggregationBuilders.terms("player_count ").field("team");
- MaxBuilder ageAgg= AggregationBuilders.max("max_age").field("age");
- sbuilder.addAggregation(teamAgg.subAggregation(ageAgg));
- SearchResponse response = sbuilder.execute().actionGet();
4.对多个field求max/min/sum/avg
例如要计算每个球队球员的平均年龄,同时又要计算总年薪,如果使用SQL语句,应表达如下:
- select team, avg(age)as avg_age, sum(salary) as total_salary from player group by team;
ES的java api:
- TermsBuilder teamAgg= AggregationBuilders.terms("team");
- AvgBuilder ageAgg= AggregationBuilders.avg("avg_age").field("age");
- SumBuilder salaryAgg= AggregationBuilders.avg("total_salary ").field("salary");
- sbuilder.addAggregation(teamAgg.subAggregation(ageAgg).subAggregation(salaryAgg));
- SearchResponse response = sbuilder.execute().actionGet();
5.聚合后对Aggregation结果排序
例如要计算每个球队总年薪,并按照总年薪倒序排列,如果使用SQL语句,应表达如下:
- select team, sum(salary) as total_salary from player group by team order by total_salary desc;
ES的java api:
- TermsBuilder teamAgg= AggregationBuilders.terms("team").order(Order.aggregation("total_salary ", false);
- SumBuilder salaryAgg= AggregationBuilders.avg("total_salary ").field("salary");
- sbuilder.addAggregation(teamAgg.subAggregation(salaryAgg));
- SearchResponse response = sbuilder.execute().actionGet();
需要特别注意的是,排序是在TermAggregation处执行的,Order.aggregation函数的第一个参数是aggregation的名字,第二个参数是boolean型,true表示正序,false表示倒序。
6.Aggregation结果条数的问题
默认情况下,search执行后,仅返回10条聚合结果,如果想反悔更多的结果,需要在构建TermsBuilder 时指定size:
- TermsBuilder teamAgg= AggregationBuilders.terms("team").size(15);
7.Aggregation结果的解析/输出
得到response后:
- <span style="white-space:pre"> </span>Map<String, Aggregation> aggMap = response.getAggregations().asMap();
- StringTerms teamAgg= (StringTerms) aggMap.get("keywordAgg");
- Iterator<Bucket> teamBucketIt = teamAgg.getBuckets().iterator();
- while (teamBucketIt .hasNext()) {
- Bucket buck = teamBucketIt .next();
-
- String team = buck.getKey();
-
- long count = buck.getDocCount();
-
- Map subaggmap = buck.getAggregations().asMap();
-
- double avg_age= ((InternalAvg) subaggmap.get("avg_age")).getValue();
-
- double total_salary = ((InternalSum) subaggmap.get("total_salary")).getValue();
-
-
- }
8. 总结
综上,聚合操作主要是调用了SearchRequestBuilder的addAggregation方法,通常是传入一个TermsBuilder,子聚合调用TermsBuilder的subAggregation方法,可以添加的子聚合有TermsBuilder、SumBuilder、AvgBuilder、MaxBuilder、MinBuilder等常见的聚合操作。
从实现上来讲,SearchRequestBuilder在内部保持了一个私有的 SearchSourceBuilder实例, SearchSourceBuilder内部包含一个List<AbstractAggregationBuilder>,每次调用addAggregation时会调用 SearchSourceBuilder实例,添加一个AggregationBuilder。
同样的,TermsBuilder也在内部保持了一个List<AbstractAggregationBuilder>,调用addAggregation方法(来自父类addAggregation)时会添加一个AggregationBuilder。有兴趣的读者也可以阅读源码的实现。
如果有任何问题,欢迎一起讨论,如果文中有什么错误,欢迎批评指正。
注:文中使用的Elastic Search API版本为2.3.2