Get sum of record count for inner bucket key in two level term aggregation

[Baekjun-Kim]

I have records with two keyword type field

• user_id: String that identifies individual user,
• result: String such as "success", "failure" or "pending" etc.

These are what I want to do:

1. Get record count for each result.
2. Exclude user, who do not have at least 5 record.

I've used "terms" aggregation to filtering out user.

GET /apple/_search
{
  "size": 0,
  "aggs": {
    "users": {
      "terms": {
        "field": "user_id.keyword",
        "min_doc_count": 5
      },
      "aggs": {
        "results": {
          "terms": {
            "field": "result.keyword",
            "size": 10
          }
        }
      }
    }
  }
}

And the result is looks like below:

{
  "aggregations": {
    "users": {
      "buckets": [
        {
          "key": "user_8",
          "doc_count": 9,
          "results": {
            "buckets": [
              {
                "key": "success",
                "doc_count": 4
              },
              {
                "key": "failure",
                "doc_count": 3
              },
              {
                "key": "pending",
                "doc_count": 2
              }
            ]
          }
        },
        {
          "key": "user_12",
          "doc_count": 7,
          "results": {
            "buckets": [
              {
                "key": "success",
                "doc_count": 2
              },
              {
                "key": "failure",
                "doc_count": 2
              },
              {
                "key": "pending",
                "doc_count": 2
              }
            ]
          }
        }
      }
    }
  }
}

Since the result, count of each "result", is calculated per user, I have to sum them.
For example, the overall summation of "success" record is 4+2 = 6. "failure" is 3+2 = 5 and so go on and on.

I think I can add "sum_bucket" aggregation to acheive it, and it works.

    "success-total": {
      "sum_bucket": {
        "buckets_path": "users>results['success']._count"
      }
    },

However the "result" field can have another string, so I can't use static string such as "success" and "pending" for it.
How can I make it to travel all bucket for each users, and get the count of each result, and result it?