如何在Elasticsearch的test_nested索引中检索指定前缀的唯一名称

Alright, let's tackle this problem step by step. You want to retrieve all unique names from your test_nested index that start with the prefix wi, and get exactly [Microsoft Windows, Widows 10, Windows XP] as the result. Since you've already checked the index mappings and run basic search requests, let's dive into the correct approach.

1. Verify Your Field Mapping First #

First off, let's confirm your name field's mapping structure—this is critical for getting accurate unique results. If you ran GET test_nested/_mappings, you’ll want to check if your name field has a keyword sub-field.

Why? Because using a text field directly for aggregation will return split, tokenized terms (not full unique names). A keyword sub-field preserves the full string, perfect for fetching unique values. Your ideal mapping should look something like this:

{
  "test_nested": {
    "mappings": {
      "properties": {
        "name": {
          "type": "text",
          "fields": {
            "keyword": {
              "type": "keyword",
              "ignore_above": 256
            }
          }
        }
      }
    }
  }
}

If your name field is nested inside an object (since your index is named test_nested), your mapping might look like this instead:

{
  "test_nested": {
    "mappings": {
      "properties": {
        "products": {
          "type": "nested",
          "properties": {
            "name": {
              "type": "text",
              "fields": {
                "keyword": {
                  "type": "keyword"
                }
              }
            }
          }
        }
      }
    }
  }
}

2. Build the Correct Search Query #

We’ll combine a prefix query to match names starting with wi with a terms aggregation to fetch unique values. Here are the two scenarios:

Scenario A: name is a Top-Level Field #

Run this search request—we set size: 0 because we don’t need raw documents, just the aggregation results:

GET test_nested/_search
{
  "size": 0,
  "query": {
    "prefix": {
      "name": "wi" # Matches any name where the tokenized text starts with "wi"
    }
  },
  "aggs": {
    "unique_matching_names": {
      "terms": {
        "field": "name.keyword", # Uses the keyword sub-field to get full unique names
        "size": 10 # Adjust this to ensure you capture all matching results
      }
    }
  }
}

Scenario B: name is a Nested Field #

If your name lives inside a nested object (like the products field above), you need to wrap the query and aggregation in a nested context:

GET test_nested/_search
{
  "size": 0,
  "query": {
    "nested": {
      "path": "products",
      "query": {
        "prefix": {
          "products.name": "wi"
        }
      }
    }
  },
  "aggs": {
    "nested_products_context": {
      "nested": {
        "path": "products"
      },
      "aggs": {
        "unique_matching_names": {
          "terms": {
            "field": "products.name.keyword",
            "size": 10
          }
        }
      }
    }
  }
}

3. Handle Case Insensitivity (Optional) #

If you need the prefix match to ignore case (e.g., match Wi, WI, or wi), add a lowercase normalizer to your keyword sub-field:

First, update the index settings:

PUT test_nested/_settings
{
  "analysis": {
    "normalizer": {
      "lowercase_normalizer": {
        "type": "custom",
        "filter": ["lowercase"]
      }
    }
  }
}

Then update the name field mapping (you’ll need to reindex if the index already has data):

PUT test_nested/_mappings
{
  "properties": {
    "name": {
      "type": "text",
      "fields": {
        "keyword": {
          "type": "keyword",
          "normalizer": "lowercase_normalizer"
        }
      }
    }
  }
}

Now your prefix query with wi will match all case variations.

4. Extract the Results #

After running the correct search, look for the aggregations.unique_matching_names.buckets array in the response. Each bucket’s key will be one of your unique names. Extracting these keys will give you exactly the list you need: [Microsoft Windows, Widows 10, Windows XP].

内容的提问来源于stack exchange，提问作者Utsav Dusad