Sorting all hits by a nested field value

Elastic Stack Elasticsearch
1

Hi!

Is it possible to sort a query result by the value of a nested field, across all hits?

To give you an example, let's say I am storing sellers and their products. A document stores the information of a single seller, including a list of their products as nested documents. So the mapping would be something like this:

{
  "mapping": {
    "properties": {
      "name": {
        "type": "text"
      },
      "products": {
        "type": "nested",
        "properties": {
          "name": {
            "type": "text"
          },
          "price": {
            "type": "integer"
          }
        }
      }
    }
  }
}

My question is, how do I retrieve all products and their vendors, listed from most expensive to cheapest? If I were working with a relational database, products and sellers would be stored in separate tables and my SQL query would look like this:

SELECT products.name, products.price, sellers.name
FROM products INNER JOIN sellers ON products.seller_id=sellers.id
ORDER BY products.price DESC;

Is it possible to achieve the same thing in Elasticsearch, with the mapping shown above? I am aware of nested queries, but so far I have only been able to sort the inner hits within each document. So I can get a result like this:

Seller A
	- Product 3, $100
	- Product 1, $80
	- Product 2, $10
Seller B
	- Product 5, $90
	- Product 4, $5

But what I need is this:

Product 3, Seller A, $100
Product 5, Seller B, $90
Product 1, Seller A, $80
Product 2, Seller A, $10
Product 4, Seller B, $5

Is this possible? Any help is greatly appreciated.

2

Why not store it as 5 separate documents (as in the results you are looking for) instead of using a single nested one?

3

Thanks for the suggestion! In the simple example I have given, this would indeed be feasible. In my
actual use-case however, "flattening" the mapping would introduce a huge redundancy. I have hundreds of fields, several of them are nested or vector-valued. (Let's say the sellers not only have names, but also addresses, phone numbers, etc... repeating all of these for every product would really blow up the size of the index.)