ElasticSearch中的一些聚合操作

今天有一個小工作，原來數據是在Hive中存着，每次查詢的時候就是後端拼接SQL然後用JDBC去數據庫中查詢，速度很慢很慢，於是我決定把數據源改爲ElasticSearch中，下面就是整個流程：

注：ElasticSearch使用的是6.3.1版本

原拼接的SQL：

select sum(cast(bdp_user_rfm.order_num as int)) as order_num,
        bdp_user_rfm.rfm_label as rfm,
        count(bdp_user_rfm.user_id) as user_count, 
        sum(cast(bdp_user_rfm.monetary as double)) as money_sum
from bdp_user_rfm inner join user_basic_info
on user_basic_info.user_id = bdp_user_rfm.user_id 
where 1=1 and bdp_user_rfm.etl_date = (select etl_date from bdp_user_rfm group by etl_date order by etl_date desc limit 1) 
and bdp_user_rfm.time = 'all' 
group by bdp_user_rfm.rfm_label;

上面用到了2張表，而且這兩張表join條件是user_id，這兩張表都有一個特點就是同一個user_id只在表中有一條記錄（就是在關係型數據庫中user_id完全可以作爲主鍵），所以我用Hive On Es把兩張表的數據跑到同一個Index中,用user_id作爲Index的id，這樣在往Es中寫數據的時候已經實現了join關聯。
上述SQL可以轉換爲如下：

ElasticSearch命令行格式：

POST /xxxxxxxxxxxxxx
{
  "size": 1,
  "aggs": {
    "group_by_rfm_label": {
      "terms": {
        "field": "rfm_label.keyword",
        "size": 10,
        "min_doc_count": 1,
        "shard_min_doc_count": 0,
        "show_term_doc_count_error": false,
        "order": [
          {
            "_count": "desc"
          },
          {
            "_key": "asc"
          }
        ]
      },
      "aggregations": {
        "order_num": {
          "sum": {
            "field": "order_num",
            "format": "#"
          }
        },
        "monetary": {
          "sum": {
            "field": "monetary"
          }
        }
      }
    }
  }
}

在接口中用Java-High-Level-Client實現：

 public static JSONArray getRFMInfo(String time, String clusterId, String key, String code) {
        JSONArray jsonArray = new JSONArray();

        RestHighLevelClient client = null;

        try {

            client = getClient();
            SearchRequest searchRequest = new SearchRequest("bdp_dev_profile_user_basic_label");
            SearchSourceBuilder sourceBuilder = new SearchSourceBuilder();

            // 時間限制 必須的字段
            BoolQueryBuilder boolQueryBuilder = QueryBuilders.boolQuery()
                    .must(QueryBuilders.termQuery("time", time));

            // clusterId 限制
            if (StringUtils.isNotBlank(clusterId)) {
                boolQueryBuilder = boolQueryBuilder.must(QueryBuilders.termQuery("clusterid.keyword", clusterId));
            }

            // 用戶組織機構過濾 使用模糊查詢
            if (StringUtils.isNotBlank(key) && StringUtils.isNotBlank(code)) {
                boolQueryBuilder = boolQueryBuilder.must(QueryBuilders.matchQuery(key, code));
            }

            MaxAggregationBuilder maxEtlDate = AggregationBuilders
                    .max("max_etl_date").field("etl_date").format("yyyy-MM-dd");

            String maxDate = "";
            try {
                ParsedMax max_etl_date = (ParsedMax) client.search(searchRequest.source(
                        new SearchSourceBuilder().aggregation(maxEtlDate)))
                        .getAggregations().asMap().get("max_etl_date");
                maxDate = max_etl_date.getValueAsString();
                System.out.println("maxDate = " + maxDate);
            } catch (Exception e) {
                e.printStackTrace();
            }

            if (StringUtils.isNotBlank(maxDate)) {
                boolQueryBuilder = boolQueryBuilder.must(QueryBuilders.termQuery("etl_date", maxDate));
            }

            TermsAggregationBuilder group_by = AggregationBuilders.terms("group_by_rfm_label").field("rfm_label.keyword");
            SumAggregationBuilder sum_order_num = AggregationBuilders.sum("order_num").field("order_num");
            //取整數
            sum_order_num.format("#");
            SumAggregationBuilder sum_monetary = AggregationBuilders.sum("monetary").field("monetary");
            //小數點後保留N位
            sum_monetary.format("0.0000000");

            group_by.subAggregation(sum_order_num);
            group_by.subAggregation(sum_monetary);

            sourceBuilder.query(boolQueryBuilder).aggregation(group_by);
            searchRequest.source(sourceBuilder);
            SearchResponse response = client.search(searchRequest);
            Map<String, Aggregation> aggregationMap = response.getAggregations().asMap();
            ParsedStringTerms user_count = (ParsedStringTerms) aggregationMap.get("group_by_rfm_label");
            for (Terms.Bucket bucket : user_count.getBuckets()) {
                JSONObject jsonObject = new JSONObject();
                jsonObject.put("rfm", bucket.getKey());
                jsonObject.put("user_count", bucket.getDocCount());
                /*一般保持客戶-->96844 206150932.24 order_num  ===>96887*/
//                System.out.println(bucket.getKey() + "-->" + bucket.getDocCount());
                Aggregations aggregations = bucket.getAggregations();
                Map<String, Aggregation> map = aggregations.asMap();
                jsonObject.put("money_sum", ((ParsedSum) map.get("monetary")).getValue());
//                System.out.println(((ParsedSum) map.get("monetary")).getValueAsString());
                jsonObject.put("order_num", ((ParsedSum) map.get("order_num")).getValue());
//                System.out.println("order_num  ===>" + ((ParsedSum) map.get("order_num")).getValueAsString());
                jsonArray.add(jsonObject);
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            returnClient(client);
        }

        return jsonArray;
    }

關於數據格式的轉換可以看我的另外一篇文章：
https://blog.csdn.net/qq_26502245/article/details/106197146