[BUG] Sum aggregation returns 0 which all documents have null values for the field

[gaobinlong]

Describe the bug

When performing sum aggregation on a field which doesn't exist in all the documents or all documents have explicit null value for that field, the aggregation result shows 0, which is not as expected, null makes more sense. And you can see that, the avg/max/min aggregation returns null.

Request:

GET test1/_search
{
  "size": 0,
  "aggs": {
    "stats-agg": {
      "stats": {
        "field": "a"
      }
    }
  }
}

result:

{
  "took": 4,
  "timed_out": false,
  "_shards": {
    "total": 1,
    "successful": 1,
    "skipped": 0,
    "failed": 0
  },
  "hits": {
    "total": {
      "value": 3,
      "relation": "eq"
    },
    "max_score": null,
    "hits": []
  },
  "aggregations": {
    "stats-agg": {
      "count": 0,
      "min": null,
      "max": null,
      "avg": null,
      "sum": 0
    }
  }
}

Related component

Search:Aggregations

To Reproduce

1. DELETE test1

PUT test1/_mapping
{
  "properties":{
    "a":{
      "type":"integer"
    }
  }
}

POST test1/_doc?refresh
{
  "a":null
}

4. Check the result

GET test1/_search
{
  "size": 0,
  "aggs": {
    "stats-agg": {
      "stats": {
        "field": "a"
      }
    }
  }
}

Expected behavior

Sum aggregation returns null rather than 0 for the case that all documents have null values for the field the aggregation performs on.

Additional Details

No response

[sandeshkr419]

[Search Triage] The sum for an empty set will be 0 - for max, min, avg - you can't have a deterministic value when your set is empty or you don't have any data, but your sum will be initialized as 0 generally.

Why do you think null makes more sense?

[gaobinlong]

Why do you think null makes more sense?

I think 0 is not able to help distinguish the case of all documents have null values for the aggregation field and the case of the actual sum equals to 0. And in most databases like mysql, sqlserver and postgresql, sum null values returns null rather than 0.