Further aggregate date histogram in Elasticsearch query

Question

I'm trying to extract, for each event_type, in how many days the event occurred and the last occurrence. Currently I wrote the following query:

{
    "size": 0,
    "query": { ... },
    "_source": false,
    "aggregations": {
        "events": { 
            "terms": { "field": "event_type" },
            "aggs": {
                "event_timestamp": {
                    "date_histogram": {
                        "field": "@timestamp",
                        "calendar_interval": "day",
                        "min_doc_count": 1,
                        "order": { "_key": "desc" }
                    }
                }
            }
        }
    }
}

that extracts data to be further transformed. In my case, I just need, for each event_type, the key of the first event_timestamp bucket and the number of event_timestamp buckets. This can be done easily in a second step after this extraction, but it would be cleaner to let Elastic do it.

Max bucket (or metric) aggregation does not seems helpful, as I don't see a suitable buckets_path in the date_histogram aggregation.

Answer 1

Good start!!

Here is a query that returns you the latest timestamp per event_type as well as the number of daily buckets for each event_type (for this you can use the bucket_script pipeline aggregation which simply returns the number of buckets in the date_histogram aggregation).

Also note that the filter_path query string parameter allows you to only return what matters to you, i.e. you can get rid of the date histogram that you're not interested in:

POST test/_search?filter_path=**.latest_timestamp.hits.hits._source,**.bucket_count,**.events.buckets.key
{
  "size": 0,
  "_source": false,
  "query": {...},
  "aggregations": {
    "events": {
      "terms": {
        "field": "event_type"
      },
      "aggs": {
        "latest_timestamp": {
          "top_hits": {
            "size": 1,
            "sort": {"@timestamp": "desc"}, 
            "_source": ["@timestamp"]
          }
        },
        "event_timestamp": {
          "date_histogram": {
            "field": "@timestamp",
            "calendar_interval": "day",
            "min_doc_count": 1,
            "order": {
              "_key": "desc"
            }
          }
        },
        "bucket_count": {
          "bucket_script": {
            "buckets_path": {
              "count": "event_timestamp._bucket_count"
            },
            "script": "params.count"
          }
        }
      }
    }
  }
}

You'll get for each event type something that looks like this:

    {
      "key": "2",                            <--- event type
      "latest": {
        "hits": {
          "hits": [
            {
              "_source": {
                "@timestamp": "2023-01-05"   <--- latest timestamp
              }
            }
          ]
        }
      },
      "bucket_count": {
        "value": 4                           <--- bucket count
      }
    },