Function score query with decay function on huge dataset

[minhdao]

Good day everyone.

Background

I'm building a search engine for a marketplace. Our market has products, sellers, and customers. Each product has many sellers with their locations. The number of sellers might be around 100k. I store seller's locations in each products (nested field and parent-child don't meet up my requirements), for e.g:

{
  "mappings": {
    "properties": {
      "stores": {
        "location": {
          "type": "geo_point"
        }
      }
    }
  }
}

When searching for products, I use function score query with decay function to re-ranking based on the distance from customer to seller's locations. score_mode is also set to max to rescore with closest sellers.

Problem

I have generated a sample data with 10k products, each product has 10k to 100k sellers to test. It returned result after 24 seconds.

Question

• Is there any way to improve this function score query?
• In case there is no way to improve, what is the practical way to do for my use case?

Thank you very much!