Strange aggregation behavior (returning partial data)

[00a45f9a00ca005db54c (Александр Кобрин)]

Hello there

Elastic aggregation returns sometimes partial data, sometimes full data from the same documents depending on date range.
Ex. docs:

"message": {
      "metadata": {
        "kitId": "07e102c2-c92c-4a9a-9d95-a7f89d006edd",
        "contentId": "1328"
      },
      "type": "APP_METRICS",
      "elasticsearch": {
        "index": "2019.03.28",
        "timestamp": "2019-03-28T14:13:49.177Z"
      },
      "payload": {
        "metric": {
          "name": "What do you expect to see here (test)?",
          "id": "b4ff6780-5163-11e9-9663-5d217198464c",
          "value": 1,
          "subname": "Fitness User"
        }
      }
    }

Multiple logs with fields "name", "subname", "value"
All of them generated with small time difference (ex. all from 28 march 2-3pm)

Aggregations are being made using mseach endpoing. Something like this:

"aggs": {
    "metrics": {
      "terms": {
        "field": "message.payload.metric.name.keyword",
        "size": 10000
      },
      "aggs": {
        "submetrics": {
          "terms": {
            "field": "message.payload.metric.subname.keyword"
          }
        },
        "data": {
          "stats": {
            "field": "message.payload.metric.value"
          }
        }
      }
    }
  }

And if I take timestamp range for example from 1553731200000 (0.00 28 march) to 1553806740000 - I will get result with sub-aggregation:

[
    {
        "name": "What do you expect to see here (test)?",
        "data": {
            "count": 8,
            "min": 1,
            "max": 1,
            "avg": 1,
            "sum": 8
        },
        "submetrics": [
            {
                "name": "Apparel Store!",
                "count": 5
            },
            {
                "name": "Children’s Store",
                "count": 1
            },
            {
                "name": "Cosmetics Store",
                "count": 1
            },
            {
                "name": "Fitness User",
                "count": 1
            }
        ]
    }
]

But if I extend "from" time to something less than 1553731200000 (ex. 1553731100000) - I'd expect to see at least same result. But I receive SAME(!) documents without subaggregation:

[
    {
        "name": "What do you expect to see here (test)?",
        "data": {
            "count": 8,
            "min": 1,
            "max": 1,
            "avg": 1,
            "sum": 8
        }
    }
]

Struggling already for 2 days.
Any ideas why is this happening? Thanks in advance

[Ignacio_Vera]

Could you provide a reproducible example with mappings and data? It is hard to understand what is going on otherwise.