Is it possible to compute "distinct sum" and "distinct average" in elasticsearch?

Question

How can I calculate a "distinct average" in elasticsearch? I have some denormalized data like this:

{ "record_id" : "100", "cost" : 42 }
{ "record_id" : "200", "cost" : 67 }
{ "record_id" : "200", "cost" : 67 }
{ "record_id" : "200", "cost" : 67 }
{ "record_id" : "400", "cost" : 11 }
{ "record_id" : "400", "cost" : 11 }
{ "record_id" : "500", "cost" : 10 }
{ "record_id" : "600", "cost" : 99 }

Notice how the "cost" is always the same for a given "record_id".

So with the above data:

1. How can I get the AVERAGE values for the "cost" field but DISTINCT by "record_id"? Result would be (42+67+11+10+99)/5=45.8

2. How can I get the SUM values for the "cost" field but DISTINCT by "record_id"? Result would be 42+67+11+10+99=229

Could I use a combination of a "terms" aggregation and then "first" and "average" sub-aggregations? I'm thinking something like this: elasticsearch calculate average of unique values

Answer 1

It's not going to work with terms aggs. Here's what's possible using painless scripts:

Indexing -- your actual mapping may differ from the generated default (esp the .keyword part on the rec_id):

POST _bulk
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"100","cost":42}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"200","cost":67}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"200","cost":67}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"200","cost":67}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"400","cost":11}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"400","cost":11}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"500","cost":10}
{"index":{"_index":"uniques","_type":"_doc"}}
{"record_id":"600","cost":99}

Then aggregating

GET uniques/_search
{
  "size": 0,
  "aggs": {
    "terms": {
      "scripted_metric": {
        "init_script": "state.id_map = [:]; state.sum = 0.0; state.elem_count = 0.0;",
        "map_script": """
          def id = doc['record_id.keyword'].value;
          if (!state.id_map.containsKey(id)) {
            state.id_map[id] = true;
            state.elem_count++;
            state.sum += doc['cost'].value;
          }
        """,
        "combine_script": """
            def sum = state.sum;
            def avg = sum / state.elem_count;
            
            def stats = [:];
            stats.sum = sum;
            stats.avg = avg;
            
            return stats
        """,
        "reduce_script": "return states"
      }
    }
  }
}

And yielding

...
"aggregations" : {
    "terms" : {
      "value" : [
        {
          "avg" : 45.8,
          "sum" : 229.0
        }
      ]
    }
  }