Elasticsearch聚合基于过滤器计数

Question

I've worked on several facet and filter based searches in the past with Solr but I'm struggling to achieve parity with Elasticsearch.

I understand that aggregations are calculated against the results of a query or globally if no query is specified. This is fine, however I would like the counts of those aggregations to be based on the results of a filter.

In Solr this is straightforward - just specify a query and filter - but with Elasticsearch a filter has no effect on aggregates and the documentation is very confusing.

My desired output for the following query is for the suggestions bucket to be scoped to the query but the resulting counts therein to be scoped to the specified filter :

{
    "size": 0,
    "query": {
        "range": {
            "published": {
                "gte": "now-1y",
                "lt": "now"
            }
        }
    },
    "filter": {
        {
            "term": {
                "tag.id": "123"
            }
        },
        {
            "term": {
                "tag.id": "456"
            }
        },
    },
    "aggs": {
        "tags": {
            "nested": {
                "path": "tag"
            },
            "aggs": {
                "suggestions": {
                    "terms": {
                        "field": "name",
                        "size": 10,
                        "min_doc_count": 1
                    },
                    "aggs": {
                        "id": {
                            "terms": {
                                "field": "id",
                                "size": 1
                            }
                        }
                    }
                }
            }
        }
    }
}

And given the example mapping:

{
    "mappings":{
        "content":{
            "properties":{
                "id":{
                    "type":"string",
                    "index":"not_analyzed"
                },
                "title":{
                    "type":"string"
                },
                "byline":{
                    "type":"string",
                    "index":"not_analyzed"
                },
                "body":{
                    "type":"string"
                },
                "publishedDate":{
                    "type":"date",
                    "format":"dateOptionalTime"
                },
                "tag":{
                    "type":"nested",
                    "include_in_parent":true,
                    "properties":{
                        "id":{
                            "type":"integer"
                        },
                        "name":{
                            "type":"string"
                        }
                    }
                }
            }
        }
    }
}

Any help is appreciated.

Answer 1

You can get the expected results by :

1. keeping the query where it is,
2. but moving the filter part (what you have is equivalent to a post_filter actually and is only applied on the results after aggregations have run) into a filter aggregation

Basically this should work:

{
  "size": 0,
  "query": {
    "range": {
      "published": {
        "gte": "now-1y",
        "lt": "now"
      }
    }
  },
  "aggs": {
    "tags": {
      "nested": {
        "path": "tag"
      },
      "aggs": {
        "suggestions": {
          "terms": {
            "field": "tag.name",
            "size": 10,
            "min_doc_count": 1
          },
          "aggs": {
            "filtered": {
              "filter": {
                "terms": {
                  "tag.id": [
                    "123",
                    "456"
                  ]
                }
              },
              "aggs": {
                "id": {
                  "terms": {
                    "field": "tag.id",
                    "size": 1
                  }
                }
              }
            }
          }
        }
      }
    }
  }
}