8-Elasticsearch之聚合函数

8-Elasticsearch之聚合函数

一 前言

聚合函数大家都不陌生,elasticsearch中也没玩出新花样,所以,这一章相对简单,只需要记得:

  • avg
  • max
  • min
  • sum

以及各自的用法即可。先来看求平均。

二 准备数据

PUT lqz/doc/1
{
  "name":"顾老二",
  "age":30,
  "from": "gu",
  "desc": "皮肤黑、武器长、性格直",
  "tags": ["黑", "长", "直"]
}

PUT lqz/doc/2
{
  "name":"大娘子",
  "age":18,
  "from":"sheng",
  "desc":"肤白貌美,娇憨可爱",
  "tags":["白", "富","美"]
}

PUT lqz/doc/3
{
  "name":"龙套偏房",
  "age":22,
  "from":"gu",
  "desc":"mmp,没怎么看,不知道怎么形容",
  "tags":["造数据", "真","难"]
}


PUT lqz/doc/4
{
  "name":"石头",
  "age":29,
  "from":"gu",
  "desc":"粗中有细,狐假虎威",
  "tags":["粗", "大","猛"]
}

PUT lqz/doc/5
{
  "name":"魏行首",
  "age":25,
  "from":"广云台",
  "desc":"仿佛兮若轻云之蔽月,飘飘兮若流风之回雪,mmp,最后竟然没有嫁给顾老二!",
  "tags":["闭月","羞花"]
}

三 avg

现在的需求是查询fromgu的人的平均年龄。

select max(age) as my_avg

GET lqz/doc/_search
{
  "query": {
    "match": {
      "from": "gu"
    }
  },
  "aggs": {
    "my_avg": {
      "avg": {
        "field": "age"
      }
    }
  },
  "_source": ["name", "age"]
}

上例中,首先匹配查询fromgu的数据。在此基础上做查询平均值的操作,这里就用到了聚合函数,其语法被封装在aggs中,而my_avg则是为查询结果起个别名,封装了计算出的平均值。那么,要以什么属性作为条件呢?是age年龄,查年龄的什么呢?是avg,查平均年龄。

返回结果如下:

{
  "took" : 1,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 3,
    "max_score" : 0.6931472,
    "hits" : [
      {
        "_index" : "lqz",
        "_type" : "doc",
        "_id" : "4",
        "_score" : 0.6931472,
        "_source" : {
          "name" : "石头",
          "age" : 29
        }
      },
      {
        "_index" : "lqz",
        "_type" : "doc",
        "_id" : "1",
        "_score" : 0.2876821,
        "_source" : {
          "name" : "顾老二",
          "age" : 30
        }
      },
      {
        "_index" : "lqz",
        "_type" : "doc",
        "_id" : "3",
        "_score" : 0.2876821,
        "_source" : {
          "name" : "龙套偏房",
          "age" : 22
        }
      }
    ]
  },
  "aggregations" : {
    "my_avg" : {
      "value" : 27.0
    }
  }
}

上例中,在查询结果的最后是平均值信息,可以看到是27岁。

虽然我们已经使用_source对字段做了过滤,但是还不够。我不想看都有哪些数据,只想看平均值怎么办?别忘了size!

GET lqz/doc/_search
{
  "query": {
    "match": {
      "from": "gu"
    }
  },
  "aggs": {
    "my_avg": {
      "avg": {
        "field": "age"
      }
    }
  },
  "size": 0, 
  "_source": ["name", "age"]
}

上例中,只需要在原来的查询基础上,增加一个size就可以了,输出几条结果,我们写上0,就是输出0条查询结果。

查询结果如下:

{
  "took" : 8,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 3,
    "max_score" : 0.0,
    "hits" : [ ]
  },
  "aggregations" : {
    "my_avg" : {
      "value" : 27.0
    }
  }
}

查询结果中,我们看hits下的total值是3,说明有三条符合结果的数据。最后面返回平均值是27。

四 max

那怎么查最大值呢?

GET lqz/doc/_search
{
  "query": {
    "match": {
      "from": "gu"
    }
  },
  "aggs": {
    "my_max": {
      "max": {
        "field": "age"
      }
    }
  },
  "size": 0
}

上例中,只需要在查询条件中将avg替换成max即可。

返回结果如下:

{
  "took" : 1,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 3,
    "max_score" : 0.0,
    "hits" : [ ]
  },
  "aggregations" : {
    "my_max" : {
      "value" : 30.0
    }
  }
}

在返回的结果中,可以看到年龄最大的是30岁。

五 min

那怎么查最小值呢?

GET lqz/doc/_search
{
  "query": {
    "match": {
      "from": "gu"
    }
  },
  "aggs": {
    "my_min": {
      "min": {
        "field": "age"
      }
    }
  },
  "size": 0
}

最小值则用min表示。

返回结果如下:

{
  "took" : 0,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 3,
    "max_score" : 0.0,
    "hits" : [ ]
  },
  "aggregations" : {
    "my_min" : {
      "value" : 22.0
    }
  }
}

返回结果中,年龄最小的是22岁。

六 sum

那么,要是想知道它们的年龄总和是多少怎么办呢?

GET lqz/doc/_search
{
  "query": {
    "match": {
      "from": "gu"
    }
  },
  "aggs": {
    "my_sum": {
      "sum": {
        "field": "age"
      }
    }
  },
  "size": 0
}

上例中,求和用sum表示。

{
  "took" : 2,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 3,
    "max_score" : 0.0,
    "hits" : [ ]
  },
  "aggregations" : {
    "my_sum" : {
      "value" : 81.0
    }
  }
}

从返回的结果可以发现,年龄总和是81岁。

七 分组查询

现在我想要查询所有人的年龄段,并且按照15~20,20~25,25~30分组,并且算出每组的平均年龄。

分析需求,首先我们应该先把分组做出来。

GET lqz/doc/_search
{
  "size": 0, 
  "query": {
    "match_all": {}
  },
  "aggs": {
    "age_group": {
      "range": {
        "field": "age",
        "ranges": [
          {
            "from": 15,
            "to": 20
          },
          {
            "from": 20,
            "to": 25
          },
          {
            "from": 25,
            "to": 30
          }
        ]
      }
    }
  }
}

上例中,在aggs的自定义别名age_group中,使用range来做分组,field是以age为分组,分组使用ranges来做,fromto是范围,我们根据需求做出三组。

{
  "took" : 3,
  "timed_out" : false,
  "_shards" : {
    "total" : 5,
    "successful" : 5,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : 5,
    "max_score" : 0.0,
    "hits" : [ ]
  },
  "aggregations" : {
    "age_group" : {
      "buckets" : [
        {
          "key" : "15.0-20.0",
          "from" : 15.0,
          "to" : 20.0,
          "doc_count" : 1
        },
        {
          "key" : "20.0-25.0",
          "from" : 20.0,
          "to" : 25.0,
          "doc_count" : 1
        },
        {
          "key" : "25.0-30.0",
          "from" : 25.0,
          "to" : 30.0,
          "doc_count" : 2
        }
      ]
    }
  }
}

返回的结果中可以看到,已经拿到了三个分组。doc_count为该组内有几条数据,此次共分为三组,查询出4条内容。还有一条数据的age属性值是30,不在分组的范围内!

那么接下来,我们就要对每个小组内的数据做平均年龄处理。

GET lqz/doc/_search
{
  "size": 0, 
  "query": {
    "match_all": {}
  },
  "aggs": {
    "age_group": {
      "range": {
        "field": "age",
        "ranges": [
          {
            "from": 15,
            "to": 20
          },
          {
            "from": 20,
            "to": 25
          },
          {
            "from": 25,
            "to": 30
          }
        ]
      },
      "aggs": {
        "my_avg": {
          "avg": {
            "field": "age"
          }
        }
      }
    }
  }
}

上例中,在分组下面,我们使用aggsage做平均数处理,这样就可以了。

{
 "took" : 1,
 "timed_out" : false,
 "_shards" : {
   "total" : 5,
   "successful" : 5,
   "skipped" : 0,
   "failed" : 0
 },
 "hits" : {
   "total" : 5,
   "max_score" : 0.0,
   "hits" : [ ]
 },
 "aggregations" : {
   "age_group" : {
     "buckets" : [
       {
         "key" : "15.0-20.0",
         "from" : 15.0,
         "to" : 20.0,
         "doc_count" : 1,
         "my_avg" : {
           "value" : 18.0
         }
       },
       {
         "key" : "20.0-25.0",
         "from" : 20.0,
         "to" : 25.0,
         "doc_count" : 1,
         "my_avg" : {
           "value" : 22.0
         }
       },
       {
         "key" : "25.0-30.0",
         "from" : 25.0,
         "to" : 30.0,
         "doc_count" : 2,
         "my_avg" : {
           "value" : 27.0
         }
       }
     ]
   }
 }
}

在结果中,我们可以清晰的看到每组的平均年龄(my_avgvalue中)。

注意:聚合函数的使用,一定是先查出结果,然后对结果使用聚合函数做处理

小结:

  • avg:求平均
  • max:最大值
  • min:最小值
  • sum:求和

欢迎斧正,that's all

posted @ 2022-03-14 01:28  刘清政  阅读(242)  评论(0编辑  收藏  举报