Find distinct inner objects in Elasticsearch

Question

We're trying to find distinct inner objects in Elasticsearch. This would be a minimum example for our case. We're stuck with something like the following mapping (changing types or indices or adding new fields wouldn't be a problem, but the structure should remain as it is):

{
  "building": {
    "properties": {
      "street": {
        "type": "string",
        "store": "yes",
        "index": "not_analyzed"
      },
      "house number": {
        "type": "string",
        "store": "yes",
        "index": "not_analyzed"
      },
      "city": {
        "type": "string",
        "store": "yes",
        "index": "not_analyzed"
      },
      "people": {
        "type": "object",
        "store": "yes",
        "index": "not_analyzed",
        "properties": {
          "firstName": {
            "type": "string",
            "store": "yes",
            "index": "not_analyzed"
          },
          "lastName": {
            "type": "string",
            "store": "yes",
            "index": "not_analyzed"
          }
        }
      }
    }
  }
}

Assuming we have this example data:

{
  "buildings": [
    {
      "street": "Baker Street",
      "house number": "221 B",
      "city": "London",
      "people": [
        {
          "firstName": "John",
          "lastName": "Doe"
        },
        {
          "firstName": "Jane",
          "lastName": "Doe"
        }
      ]
    },
    {
      "street": "Baker Street",
      "house number": "5",
      "city": "London",
      "people": [
        {
          "firstName": "John",
          "lastName": "Doe"
        }
      ]
    },
    {
      "street": "Garden Street",
      "house number": "1",
      "city": "London",
      "people": [
        {
          "firstName": "Jane",
          "lastName": "Smith"
        }
      ]
    }
  ]
}

When we query for the street "Baker Street" (and whatever additional options needed), we expect to get the following list:

[
    {
      "firstName": "John",
      "lastName": "Doe"
    },
    {
      "firstName": "Jane",
      "lastName": "Doe"
    }
]

The format does not matter too much, but we should be able to parse the first and last name. Just, as our actual data-set is much larger, we need the entries to be distinct.

We are using Elasticsearch 1.7.

Answer 1

We finally solved our problem.

Our solution is (as we expected) a pre-calculated people_all field. But instead of using copy_to or transform we're just writing it as we are writing the other fields when importing our data. The field looks as follows:

"people": {
  "type": "nested",
  ..
  "properties": {
    "firstName": {
      "type": "string",
      "store": "yes",
      "index": "not_analyzed"
    },
    "lastName": {
      "type": "string",
      "store": "yes",
      "index": "not_analyzed"
    },
    "people_all": {
      "type": "string",
      "index": "not_analyzed"
    }
  }
}

Please pay attention on the "index": "not_analyzed" at the people_all field. This is important to have complete buckets. If you don't use it, our example will return 3 buckets "john", "jane" and "doe".

After writing this new field we can run an aggragetion as follows:

{
  "size": 0,
  "query": {
    "term": {
      "street": "Baker Street"
    }
  },
  "aggs": {
    "people_distinct": {
      "nested": {
        "path": "people"
      },
      "aggs": {
        "people_all_distinct": {
          "terms": {
            "field": "people.people_all",
            "size": 0
          }
        }
      }
    }
  }
}

And we return the following response:

{
  "took": 2,
  "timed_out": false,
  "_shards": {
    "total": 5,
    "successful": 5,
    "failed": 0
  },
  "hits": {
    "total": 2,
    "max_score": 0.0,
    "hits": []
  },
  "aggregations": {
    "people_distinct": {
      "doc_count": 3,
      "people_name_distinct": {
        "doc_count_error_upper_bound": 0,
        "sum_other_doc_count": 0,
        "buckets": [
          {
            "key": "John Doe",
            "doc_count": 2
          },
          {
            "key": "Jane Doe",
            "doc_count": 1
          }
        ]
      }
    }
  }
}

Out of the buckets in the response we are now able to create the distinct people objects.

Please let us know if there is a better way to reach our goal. Parsing the buckets is not an optimal solution and it would be more fancy to have the fields firstName and lastName in each bucket.