elasticsearch: using nested agg after reverse_nested shows higher count than expected - elasticsearch

Question

Using Elasticsearch 2.2.0, I am doing this:
Grouping by a nested field: nested_path.nested_field
Using a reverse_nested agg so I can apply this filter: non_nested_field == "yay"
Using a nested agg so I can then get a count of the nested field I am grouping by: nested_path.nested_field
Problem: By using the reverse_nested agg I am getting a higher doc_count than I would expect.
Here is the mapping and docs I am indexing:
PUT /my_index
{
"mappings": {
"my_type": {
"properties": {
"nested_path": {
"type": "nested",
"properties": {
"nested_field": {
"type": "string"
}
}
},
"non_nested_field": {
"type": "string"
}
}
}
}
}
POST /my_index/my_type/1
{
"non_nested_field": "whoray",
"nested_path": [
{
"nested_field": "yes"
},
{
"nested_field": "yes"
},
{
"nested_field": "no"
}
]
}
POST /my_index/my_type/2
{
"non_nested_field": "yay",
"nested_path": [
{
"nested_field": "maybe"
},
{
"nested_field": "no"
}
]
}
Request body:
POST my_index/my_type/_search
{
"aggs": {
"nested_option": {
"nested": {
"path": "nested_path"
},
"aggs": {
"group_list": {
"terms": {
"field": "nested_path.nested_field",
"size": 100
},
"aggs": {
"level_1": {
"reverse_nested": {},
"aggs": {
"level_2": {
"filter": {
"term": {
"non_nested_field": "yay"
}
},
"aggs": {
"level_3": {
"nested": {
"path": "nested_path"
},
"aggs": {
"stat": {
"value_count": {
"field": "nested_path.nested_field"
}
}
}
}
}
}
}
}
}
}
}
}
},
"size": 0
}
Part of the response I get is this:
{
"aggregations": {
"nested_option": {
"doc_count": 5,
"group_list": {
"buckets": [
{
"key": "no",
"doc_count": 2,
"level_1": {
"doc_count": 2,
"level_2": {
"doc_count": 1,
"level_3": {
"doc_count": 2,
"stat": {
"value": 2
}
}
}
}
}
//....
]
}
}
}
}
In the first element of the buckets array in the response, level_1.level_2.doc_count is 1, and this is correct, because there's only one of the two docs indexed where nested_path.nested_field == "no" and non_nested_field == "yay". But level_1.level_2.level_3.doc_count in the response is 2. It should only be 1. This seems like a bug to me.