Fix multi-value handling in composite agg

[ywelsch]

Multi-valued number fields are handled differently by the composite terms agg in comparison to the regular terms agg. In particular, we use set semantics when grouping on multi-valued numeric fields using terms / multi-terms aggs (see e.g.

elasticsearch/server/src/main/java/org/elasticsearch/search/aggregations/bucket/terms/NumericTermsAggregator.java

Line 96 in d1ba276

if (previous != val || i == 0) {

) while we don't do the same for composite terms aggs. The following example in particular shows some discrepancies, which are fixed in this PR:

DELETE test

PUT test/_doc/id1?refresh=true
{
  "foo" : [1, 1]
}

GET test/_search
{
  "docvalue_fields": ["foo"],
  "aggs": {
    "terms_foo": {
      "terms": {
        "field": "foo"
      },
      "aggs": {
        "stats": {
          "sum": {
            "field": "foo"
          }
        }
      }
    },
    "comp_foo": {
      "composite": {
        "sources": [
          {
            "foo": {
              "terms": {
                "field": "foo"
              }
            }
          }
        ]
      },
      "aggs": {
        "stats": {
          "sum": {
            "field": "foo"
          }
        }
      }
    }
  }
}

which yields

{
  "took": 2,
  "timed_out": false,
  "_shards": {
    "total": 1,
    "successful": 1,
    "skipped": 0,
    "failed": 0
  },
  "hits": {
    "total": {
      "value": 1,
      "relation": "eq"
    },
    "max_score": 1,
    "hits": [
      {
        "_index": "test",
        "_id": "id1",
        "_score": 1,
        "_source": {
          "foo": [
            1,
            1
          ]
        },
        "fields": {
          "foo": [
            1,
            1
          ]
        }
      }
    ]
  },
  "aggregations": {
    "comp_foo": {
      "after_key": {
        "foo": 1
      },
      "buckets": [
        {
          "key": {
            "foo": 1
          },
          "doc_count": 1,
          "stats": {
            "value": 4
          }
        }
      ]
    },
    "terms_foo": {
      "doc_count_error_upper_bound": 0,
      "sum_other_doc_count": 0,
      "buckets": [
        {
          "key": 1,
          "doc_count": 1,
          "stats": {
            "value": 2
          }
        }
      ]
    }
  }
}

Relates #88598

[elasticsearchmachine]

Hi @ywelsch, I've created a changelog YAML for you.

[ywelsch]

NumericTermsAggregator implements the same "set" semantics here.

[elasticsearchmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)

[ywelsch]

@elasticsearchmachine run elasticsearch-ci/part-1
@elasticsearchmachine run elasticsearch-ci/docs-check

[nik9000]

👍 Aggs shouldn't collect twice for two values that fall into the same bucket.

[ywelsch]

Thanks @nik9000!!!!