Discovery autocomplete feature cause fielddata of indices significant increase

[LoadingZhang]

Kibana version: 6.4.0
Elasticsearch version: 6.4.0
Describe the bug:
There are 16TB+/150+Bil indices in ES, also on fields with very high cardinality such as uid.
If turn on autocomplete feature, then input uid: in search bar, kibana would send this query to ES:

{
  "size": 0,
  "timeout": "1s",
  "terminate_after": 100000,
  "query": {
    "match_all": {
      "boost": 1.0
    }
  },
  "aggregations": {
    "suggestions": {
      "terms": {
        "field": "uid",
        "size": 10,
        "shard_size": 10,
        "min_doc_count": 1,
        "shard_min_doc_count": 0,
        "show_term_doc_count_error": false,
        "execution_hint": "map",
        "order": [
          {
            "_count": "desc"
          },
          {
            "_key": "asc"
          }
        ],
        "include": ".*"
      }
    }
  }
}

It will load 160GB+ fielddata to memory, then ES OOM.
Can we have better way to implement this suggestion function to avoid load too much fielddata memory?

[elasticmachine]

Pinging @elastic/kibana-app

[LoadingZhang]

this problem may be solved by this issue, thanks

[timroes]

@Bargs I am reopening this for further discussion, since there are reports this is not yet fixed. I am wondering why we're sending execution_hint: map, since the documentation states, that it should mainly be used when only a few documents match the query. For our KQL auto complete isn't it most often the opposite?

[elasticmachine]

Pinging @elastic/kibana-app-arch (Team:AppArch)

[lukasolson]

Closing this as resolved due to now using _terms_enum API for autocomplete suggestions.