Mean aggregation over cadinality aggregation, over previous composite aggregation

My use case require to obtain an average over previous cardinality aggregation, this aggregations works over the buckets generated by a composite aggregation "calendar histogram" that breaks the timeline in calendar days, due to I'm working over a long time span.

I don't know whats fits best to solve this case using Elasticsearch capabilities, sub-aggregation, pipeline-aggregation ... , the cardinality column is generated over the calendar buckets using sub-aggregation, I would like work over this output to obtanin an average.

aggregations: {
    "daily_product_names_for_specific_shop": {
        "composite": {
            "size": 10000,
            "sources": [
              {
                "date": {
                  "date_histogram": {
                    "field": "@timestamp",
                    "calendar_interval": "1d",
                    "format": "iso8601"
                  }
                }
              }
            ]
        },
        # Sub-aggregation over composite buckets
        "aggregations": {
           "different_products_amount": {
               "cardinality": {
                    "field": "ticket.p_name"
                }
           }
        # Maybe some stuff
        }
    }
}

In my opinion, average bucket aggregation of pipeline aggregation fits your purpose. If it doesn't work, please ask again.

I have tried:

aggregations: {
    "daily_product_names_for_specific_shop": {
        "composite": {
            "size": 10000,
            "sources": [
              {
                "date": {
                  "date_histogram": {
                    "field": "@timestamp",
                    "calendar_interval": "1d",
                    "format": "iso8601"
                  }
                }
              }
            ]
        },
        # Sub-aggregation over composite buckets
        "aggregations": {
           "different_products_amount": {
               "cardinality": {
                    "field": "ticket.p_name"
                }
           },
           # The avg
           "mean": {
                "avg_bucket": {
                    "buckets_path": "different_products_amount",
                    "gap_policy": "skip",
                    "format": "#,##0.00;(#,##0.00)"
              }
        }
    }
}

But gets:

Validation Failed: 1: The first aggregation in buckets_path must be a multi-bucket aggregation for aggregation [mean] found :org.elasticsearch.search.aggregations.metrics.CardinalityAggregationBuilder for buckets path: different_products_amount;

Also I have already tried:

aggregations: {
    "daily_product_names_for_specific_shop": {
        "composite": {
            "size": 10000,
            "sources": [
              {
                "date": {
                  "date_histogram": {
                    "field": "@timestamp",
                    "calendar_interval": "1d",
                    "format": "iso8601"
                  }
                }
              }
            ]
        },
        # Sub-aggregation over composite buckets
        "aggregations": {
           "different_products_amount": {
               "cardinality": {
                    "field": "ticket.p_name"
                }
           }
        }
    },
   "mean": {
       "avg_bucket": {
                    "buckets_path": "daily_product_names_for_specific_shop>different_products_amount",
                    "gap_policy": "skip",
                    "format": "#,##0.00;(#,##0.00)"
       }
}

But the composite acts:

Validation Failed: 1: The first aggregation in buckets_path must be a multi-bucket aggregation for aggregation [mean] found :org.elasticsearch.search.aggregations.bucket.composite.CompositeAggregationBuilder for buckets path: daily_product_names_for_specific_shop>different_products_amount;

The latter query looks correct grammarticaly. Sorry, I didn't know pipeline aggregation can not be used with composit aggregation.

I found an issue explaining the reason.

Are there any destructive problems by using only datetime histogram aggregation without composite aggregation and discard unnecessary outputs?