how to return the count of unique documents by using elasticsearch aggregation

I think you need a reverse_nested aggregation, because you want aggregation based on a nested value, but actually counting the ROOT documents, not the nested ones

{  "query": {    "bool": {      "must": [        {          "term": {            "last_name": "smith"          }        }      ]    }  },  "aggs": {    "location": {      "nested": {        "path": "location"      },      "aggs": {        "state": {          "terms": {            "field": "location.state",            "size": 10          },          "aggs": {            "top_reverse_nested": {              "reverse_nested": {}            }          }        }      }    }  }}

And, as a result, you would see something like this:

"aggregations": {      "location": {         "doc_count": 6,         "state": {            "doc_count_error_upper_bound": 0,            "sum_other_doc_count": 0,            "buckets": [               {                  "key": "ny",                  "doc_count": 4,                  "top_reverse_nested": {                     "doc_count": 2                  }               },               {                  "key": "ca",                  "doc_count": 2,                  "top_reverse_nested": {                     "doc_count": 2                  }               }            ]         }      }   }

And what you are looking for is under top_reverse_nested part.One point here: if I'm not mistaking "doc_count": 6 is the NESTED document count, so don't be confused about these numbers thinking you are counting root documents, the count is on the nested ones. So, for a document with three nested ones that match, the count would be 3, not 1.