composite agg fails to bucket date_nanos with date_histogram source

[benwtrent]

This bug is present in at least 7.6. Probably earlier.

composite agg does not behave the same as date_histogram on a date_nanos field.

Example:
Lets build a date_nanos index with some docs

PUT mah_nano
{
  "mappings": {
    "properties": {
      "@timestamp": {
        "type": "date_nanos"
      }
    }
  }
}

POST mah_nano/_doc
{
  "@timestamp": "2019-07-15T16:50:31+00:00"
}

POST mah_nano/_doc
{
  "@timestamp": "2019-07-16T16:50:31+00:00"
}

POST mah_nano/_doc
{
  "@timestamp": "2019-07-17T16:50:31+00:00"
}

POST mah_nano/_doc
{
  "@timestamp": "2019-07-17T16:55:31+00:00"
}

A good call from a date_histogram agg:

GET mah_nano/_search?size=0
{
  "aggs": {
    "buckets": {
      "date_histogram": {
        "field": "@timestamp",
        "calendar_interval": "day"
      },
      "aggs": {
        "avg": {
          "avg": {
            "field": "@timestamp"
          }
        }
      }
    }
  }
}

Returns:

"aggregations" : {
    "buckets" : {
      "buckets" : [
        {
          "key_as_string" : "2019-07-15T00:00:00.000Z",
          "key" : 1563148800000,
          "doc_count" : 1,
          "avg" : {
            "value" : 1.563209431E12,
            "value_as_string" : "2019-07-15T16:50:31.000Z"
          }
        },
        {
          "key_as_string" : "2019-07-16T00:00:00.000Z",
          "key" : 1563235200000,
          "doc_count" : 1,
          "avg" : {
            "value" : 1.563295831E12,
            "value_as_string" : "2019-07-16T16:50:31.000Z"
          }
        },
        {
          "key_as_string" : "2019-07-17T00:00:00.000Z",
          "key" : 1563321600000,
          "doc_count" : 2,
          "avg" : {
            "value" : 1.563382381E12,
            "value_as_string" : "2019-07-17T16:53:01.000Z"
          }
        }
      ]
    }

But, the following composite agg does not return the same thing:

GET mah_nano/_search?size=0
{
  "aggs": {
    "buckets": {
      "composite": {
        "sources": [
          {
            "timestamp": {
              "date_histogram": {
                "field": "@timestamp",
                "calendar_interval": "day"
              }
            }
          }
        ]
      },
      "aggs": {
        "avg": {
          "avg": {
            "field": "@timestamp"
          }
        }
      }
    }
  }
}

There are instead strange null values and bucket keys

"aggregations" : {
    "buckets" : {
      "after_key" : {
        "timestamp" : 1563382530921600000
      },
      "buckets" : [
        {
          "key" : {
            "timestamp" : 1563209430940800000
          },
          "doc_count" : 1,
          "avg" : {
            "value" : null
          }
        },
        {
          "key" : {
            "timestamp" : 1563295830940800000
          },
          "doc_count" : 1,
          "avg" : {
            "value" : null
          }
        },
        {
          "key" : {
            "timestamp" : 1563382230940800000
          },
          "doc_count" : 1,
          "avg" : {
            "value" : null
          }
        },
        {
          "key" : {
            "timestamp" : 1563382530921600000
          },
          "doc_count" : 1,
          "avg" : {
            "value" : null
          }
        }
      ]
    }
  }

[elasticmachine]

Pinging @elastic/es-analytics-geo (:Analytics/Aggregations)

[nik9000]

This is what you get with a non-nanos-date:

{
  "took" : 17,
  "timed_out" : false,
  "_shards" : {
    "total" : 1,
    "successful" : 1,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : {
      "value" : 4,
      "relation" : "eq"
    },
    "max_score" : null,
    "hits" : [ ]
  },
  "aggregations" : {
    "buckets" : {
      "after_key" : {
        "timestamp" : 1563321600000
      },
      "buckets" : [
        {
          "key" : {
            "timestamp" : 1563148800000
          },
          "doc_count" : 1,
          "avg" : {
            "value" : 1.563209431E12,
            "value_as_string" : "2019-07-15T16:50:31.000Z"
          }
        },
        {
          "key" : {
            "timestamp" : 1563235200000
          },
          "doc_count" : 1,
          "avg" : {
            "value" : 1.563295831E12,
            "value_as_string" : "2019-07-16T16:50:31.000Z"
          }
        },
        {
          "key" : {
            "timestamp" : 1563321600000
          },
          "doc_count" : 2,
          "avg" : {
            "value" : 1.563382381E12,
            "value_as_string" : "2019-07-17T16:53:01.000Z"
          }
        }
      ]
    }
  }
}

Which, I think, is pretty much what you should expect from a nanos date. I'll see if I can make it do that.