elasticsearc使用指南之ES管道聚合(Pipeline Aggregation)

管道聚合处理来自其他聚合而不是文档集的输出，将信息添加到输出树中。

注：关于脚本聚合目前在本文中暂时不会涉及。

主要有如下两种管道聚合方式：

• parent
• sibling

下面一一介绍ES定义的管道聚合。

1、Avg Bucket Aggregation

同级管道聚合，它计算同级聚合中指定度量的平均值。同级聚合必须是多桶聚合，针对的是度量聚合(metric Aggregation)。
示例如下：

{
    "avg_bucket": {
        "buckets_path": "the_sum"  // @1
    }
}

• buckets_path：指定聚合的名称，支持多级嵌套聚合。
  其他参数：
• gap_policy
  当管道聚合遇到不存在的值，有点类似于term等聚合的(missing)时所采取的策略，可选择值为：skip、insert_zeros。
• skip：此选项将丢失的数据视为bucket不存在。它将跳过桶并使用下一个可用值继续计算。
• insert_zeros：默认使用0代替。
• format
  用于格式化聚合桶的输出(key)。

示例如下：

POST /_search
{
  "size": 0,
  "aggs": {
    "sales_per_month": {                  // @1
           "date_histogram": {
            "field": "date",
            "interval": "month"
      },
      "aggs": {                                    // @2
        "sales": {
          "sum": {
            "field": "price"
          }
        }
      }
    },
    "avg_monthly_sales": {             // @3
      "avg_bucket": {
        "buckets_path": "sales_per_month>sales" 
      }
    }
  }
}

代码@1：首先定义第一级聚合（按月）直方图聚合。
代码@2：定义第二级聚合，在按月聚合的基础上，对每个月的文档求sum。
代码@3：对上面的聚合求平均值。

其返回结果如下：

{
    ... // 省略
   "aggregations": {
      "sales_per_month": {
         "buckets": [
            {
               "key_as_string": "2015/01/01 00:00:00",
               "key": 1420070400000,
               "doc_count": 3,
               "sales": {
                  "value": 550.0
               }
            },
            {
               "key_as_string": "2015/02/01 00:00:00",
               "key": 1422748800000,
               "doc_count": 2,
               "sales": {
                  "value": 60.0
               }
            }
         ]
      },
      "avg_monthly_sales": {   // 这是对二级聚合的结果再进行一次求平均值聚合。
          "value": 328.33333333333333
      }
   }
}

对应的JAVA示例如下：

public static void test_pipeline_avg_buncket_aggregation() {
        RestHighLevelClient client = EsClient.getClient();
        try {
            SearchRequest searchRequest = new SearchRequest();
            searchRequest.indices("aggregations_index02");
            SearchSourceBuilder sourceBuilder = new SearchSourceBuilder();
            AggregationBuilder aggregationBuild = AggregationBuilders.terms("seller_agg")
                                                        .field("sellerId")
                                                        .subAggregation(AggregationBuilders.sum("seller_num_agg")
                                                                            .field("num")
                                                        )
                                                  ;
            sourceBuilder.aggregation(aggregationBuild);
            
            // 添加 avg bucket pipeline
            sourceBuilder.aggregation(new AvgBucketPipelineAggregationBuilder("seller_num_agg_av", "seller_agg>seller_num_agg"));
            sourceBuilder.size(0);
            
            searchRequest.source(sourceBuilder);
            SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT);
            System.out.println(result);
        } catch (Throwable e) {
            e.printStackTrace();
        } finally {
            EsClient.close(client);
        }
    }

2、Percentiles Bucket Aggregation

同级管道聚合，百分位管道聚合。其JAVA示例如下：

public static void test_Percentiles_buncket_aggregation() {
        RestHighLevelClient client = EsClient.getClient();
        try {
            SearchRequest searchRequest = new SearchRequest();
            searchRequest.indices("aggregations_index02");
            SearchSourceBuilder sourceBuilder = new SearchSourceBuilder();
            AggregationBuilder aggregationBuild = AggregationBuilders.terms("seller_agg")
                                                        .field("sellerId")
                                                        .subAggregation(AggregationBuilders.sum("seller_num_agg")
                                                                            .field("num")
                                                        )
                                                  ;
            sourceBuilder.aggregation(aggregationBuild);
            
            // 添加 avg bucket pipeline
            sourceBuilder.aggregation(new PercentilesBucketPipelineAggregationBuilder("seller_num_agg_av", "seller_agg>seller_num_agg"));
            sourceBuilder.size(0);
            
            searchRequest.source(sourceBuilder);
            SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT);
            System.out.println(result);
        } catch (Throwable e) {
            e.printStackTrace();
        } finally {
            EsClient.close(client);
        }
    }

其返回值如下：

{
  ...  // 省略其他属性
    "aggregations":{
        "lterms#seller_agg":{
            "doc_count_error_upper_bound":0,
            "sum_other_doc_count":12,
            "buckets":[
                {
                    "key":45,
                    "doc_count":567,
                    "sum#seller_num_agg":{
                        "value":911
                    }
                },
                {
                    "key":31,
                    "doc_count":324,
                    "sum#seller_num_agg":{
                        "value":353
                    }
                } // 省略其他桶的显示
            ]
        },
        "percentiles_bucket#seller_num_agg_av":{
            "values":{
                "1.0":5,
                "5.0":5,
                "25.0":10,
                "50.0":20,
                "75.0":290,
                "95.0":911,
                "99.0":911
            }
        }
    }
}

3、Cumulative Sum Aggregation

累积管道聚合，就是就是依次将每个管道的sum聚合进行累加。

其语法(restfull)如下：

{
    "cumulative_sum": {
        "buckets_path": "the_sum"
    }
}

支持的参数说明：

• buckets_path
  桶聚合名称，作为管道聚合的输入信息。
• format
  格式化key。

使用示例如下：

POST /sales/_search
{
    "size": 0,
    "aggs" : {
        "sales_per_month" : {
            "date_histogram" : {
                "field" : "date",
                "interval" : "month"
            },
            "aggs": {
                "sales": {
                    "sum": {
                        "field": "price"
                    }
                },
                "cumulative_sales": {
                    "cumulative_sum": {
                        "buckets_path": "sales" 
                    }
                }
            }
        }
    }
}

其返回结果如下：

{
   "took": 11,
   "timed_out": false,
   "_shards": ...,
   "hits": ...,
   "aggregations": {
      "sales_per_month": {
         "buckets": [
            {
               "key_as_string": "2015/01/01 00:00:00",
               "key": 1420070400000,
               "doc_count": 3,
               "sales": {
                  "value": 550.0
               },
               "cumulative_sales": {
                  "value": 550.0
               }
            },
            {
               "key_as_string": "2015/02/01 00:00:00",
               "key": 1422748800000,
               "doc_count": 2,
               "sales": {
                  "value": 60.0
               },
               "cumulative_sales": {
                  "value": 610.0
               }
            },
            {
               "key_as_string": "2015/03/01 00:00:00",
               "key": 1425168000000,
               "doc_count": 2,
               "sales": {
                  "value": 375.0
               },
               "cumulative_sales": {
                  "value": 985.0
               }
            }
         ]
      }
   }
}

从结果可知，cumulative_sales的值等于上一个cumulative_sales + 当前桶的sum聚合。

对应的JAVA示例如下：

{
    "aggregations":{
        "date_histogram#createTime_histogram":{
            "buckets":{
                "2015-12-01 00:00:00":{
                    "key_as_string":"2015-12-01 00:00:00",
                    "key":1448928000000,
                    "doc_count":6,
                    "sum#seller_num_agg":{
                        "value":16
                    },
                    "simple_value#Cumulative_Seller_num_agg":{
                        "value":16
                    }
                },
                "2016-01-01 00:00:00":{
                    "key_as_string":"2016-03-01 00:00:00",
                    "key":1456790400000,
                    "doc_count":10,
                    "sum#seller_num_agg":{
                        "value":11
                    },
                    "simple_value#Cumulative_Seller_num_agg":{
                        "value":31
                    }
                }
                // ... 忽略
            }
        }
    }
}

4、Bucket Sort Aggregation

一种父管道聚合，它对其父多桶聚合的桶进行排序。并可以指定多个排序字段。每个bucket可以根据它的_key、_count或子聚合进行排序。此外，可以设置from和size的参数，以便截断结果桶。

使用语法如下：

{
    "bucket_sort": {
        "sort": [
            {"sort_field_1": {"order": "asc"}},
            {"sort_field_2": {"order": "desc"}},
            "sort_field_3"
        ],
        "from": 1,
        "size": 3
    }
}

支持的参数说明如下：

• sort
  定义排序结构。
• from
  用与对父聚合的桶进行截取，该值之前的所有桶将忽略，也就是不参与排序，默认为0。
• size
  返回的桶数。默认为父聚合的所有桶。
• gap_policy
  当管道聚合遇到不存在的值，有点类似于term等聚合的(missing)时所采取的策略，可选择值为：skip、insert_zeros。
• skip：此选项将丢失的数据视为bucket不存在。它将跳过桶并使用下一个可用值继续计算。
• insert_zeros：默认使用0代替。

官方示例如下：

POST /sales/_search
{
    "size": 0,
    "aggs" : {
        "sales_per_month" : {
            "date_histogram" : {
                "field" : "date",
                "interval" : "month"
            },
            "aggs": {
                "total_sales": {
                    "sum": {
                        "field": "price"
                    }
                },
                "sales_bucket_sort": {
                    "bucket_sort": {
                        "sort": [
                          {"total_sales": {"order": "desc"}}
                        ],
                        "size": 3
                    }
                }
            }
        }
    }
}

对应的JAVA示例如下：

public static void test_bucket_sort_Aggregation() {
        RestHighLevelClient client = EsClient.getClient();
        try {
            
            //构建日期直方图聚合  时间间隔，示例中按月统计
            DateHistogramInterval interval = new DateHistogramInterval("1M"); 
            SearchRequest searchRequest = new SearchRequest();
            searchRequest.indices("aggregations_index02");
            SearchSourceBuilder sourceBuilder = new SearchSourceBuilder();
            AggregationBuilder aggregationBuild = AggregationBuilders.dateHistogram("createTime_histogram")
                                                                        .field("createTime")
                                                                        .dateHistogramInterval(interval)
                                                                        .keyed(true)
                                                                        .subAggregation(AggregationBuilders.sum("seller_num_agg")
                                                                                .field("num")
                                                                        )
                                                                        .subAggregation(new  BucketSortPipelineAggregationBuilder("seller_num_agg_sort", Arrays.asList(
                                                                                new FieldSortBuilder("seller_num_agg").order(SortOrder.ASC)))
                                                                                .from(0)
                                                                                .size(3))
                                                                        //  BucketSortPipelineAggregationBuilder(String name, List<FieldSortBuilder> sorts)
                                                                        .subAggregation(new CumulativeSumPipelineAggregationBuilder("Cumulative_Seller_num_agg", "seller_num_agg"))
                                                                    //    .format("yyyy-MM-dd") // 对key的格式化
                                                  ;
            sourceBuilder.aggregation(aggregationBuild);
            sourceBuilder.size(0);
            sourceBuilder.query(
                    QueryBuilders.termQuery("sellerId", 24)
            );
            searchRequest.source(sourceBuilder);
            SearchResponse result = client.search(searchRequest, RequestOptions.DEFAULT);
            System.out.println(result);
        } catch (Throwable e) {
            e.printStackTrace();
        } finally {
            EsClient.close(client);
        }
    }

返回值：

{
    "aggregations":{
        "date_histogram#createTime_histogram":{
            "buckets":{
                "2016-04-01 00:00:00":{
                    "key_as_string":"2016-04-01 00:00:00",
                    "key":1459468800000,
                    "doc_count":2,
                    "sum#seller_num_agg":{
                        "value":2
                    },
                    "simple_value#Cumulative_Seller_num_agg":{
                        "value":2
                    }
                },
                "2017-05-01 00:00:00":{
                    "key_as_string":"2017-05-01 00:00:00",
                    "key":1493596800000,
                    "doc_count":3,
                    "sum#seller_num_agg":{
                        "value":3
                    },
                    "simple_value#Cumulative_Seller_num_agg":{
                        "value":5
                    }
                },
                "2017-02-01 00:00:00":{
                    "key_as_string":"2017-02-01 00:00:00",
                    "key":1485907200000,
                    "doc_count":4,
                    "sum#seller_num_agg":{
                        "value":4
                    },
                    "simple_value#Cumulative_Seller_num_agg":{
                        "value":9
                    }
                }
            }
        }
    }

5、Max Bucket Aggregation

与 avg类似。

6、Min Bucket Aggregation

与 avg类似。

7、Sum Bucket Aggregation

与 avg类似。

8、Stats Bucket Aggregation

与 avg类似。

本节详细介绍了ES Pipeline Aggregation 管道聚合的使用方法，重点介绍了Avg Bucket Aggregation、Percentiles Bucket Aggregation、Cumulative Sum Aggregation、Bucket Sort Aggregation、Max Bucket Aggregation、Min Bucket Aggregation、Sum Bucket Aggregation、Stats Bucket Aggregation。

原文发布时间为：2019-03-18
本文作者：丁威，《RocketMQ技术内幕》作者。
本文来自中间件兴趣圈，了解相关信息可以关注中间件兴趣圈。