Elasticsearch - combine fields from multiple documents

You can use Scripted Metric Aggregation with a reduce_script.

Setup some test data:

curl -XPUT http://localhost:9200/testing/foo/1 -d '{ "foo" : [1, 2, 3] }'curl -XPUT http://localhost:9200/testing/foo/2 -d '{ "foo" : [4, 5, 6] }'

Now try this aggregation:

curl -XGET "http://localhost:9200/testing/foo/_search" -d'{  "size": 0,  "aggs": {    "fooreduced": {      "scripted_metric": {        "init_script": "_agg[\"result\"] = []",        "map_script":  "_agg.result.add(doc[\"foo\"].values)",        "reduce_script": "reduced = []; for (a in _aggs) { for (entry in a) { word = entry.key; reduced += entry.value } }; return reduced.flatten().sort()"      }    }  }}'

The call will return this:

{  "took": 50,  "timed_out": false,  "_shards": {    "total": 6,    "successful": 6,    "failed": 0  },  "hits": {    "total": 2,    "max_score": 0,    "hits": []  },  "aggregations": {    "fooreduced": {      "value": [        1,        2,        3,        4,        5,        6      ]    }  }}

It might be possible that there is a solution withoun .flatten(), but I'm not that much into groovy (yet) to find such a solution. And I can't say how good the performance of this aggregation is, you have to test it for yourself.