Вложенные поля и поля-объекты

Real-world documents often contain nested structures: a blog post with comments, a product with variants, or an order with line items. Elasticsearch must decide how to index these JSON objects so they stay searchable.

This lesson covers the two main approaches: the default object type and the specialized nested type.

By default, any JSON object inside a document is mapped as the object type. Elasticsearch flattens the inner fields into dotted paths.

A field author.name simply becomes a normal Lucene field. This is efficient and works perfectly for single objects.

Internally the object above is stored as two flat fields: author.first = Jane and author.last = Doe. The hierarchy is only conceptual; Lucene sees independent fields.

This is fine until you have an array of objects.

Consider an array of users. After flattening, Elasticsearch loses the link between which first name belongs to which last name.

The arrays become user.first = [Alice, John] and user.last = [White, Smith] separately.

A query for first = Alice AND last = Smith would incorrectly match the document above, because the cross-object relationship is gone. The values are pooled together.

The nested type solves this.

Set the field type to nested in the mapping. Each object in the array is then indexed as a hidden, separate Lucene document, preserving its internal field relationships.

You must use a nested query and specify the path. Conditions inside are evaluated against a single sub-document, so cross-object false matches disappear.

Add inner_hits to a nested query to return which specific sub-document(s) matched, not just the parent document. This is essential for highlighting the relevant array element.

Nested fields are powerful but have trade-offs:

• Each array element is a separate Lucene doc, increasing index size.
• Updating one element re-indexes the whole parent document.
• Deeply nested or large arrays can hurt performance.

Use the index.mapping.nested_objects.limit setting to cap counts.

For tightly coupled data that updates together, nested is ideal. For independently updated, high-cardinality relationships, consider the join (parent-child) field type instead, which decouples updates at a higher query cost.

Use object when arrays do not require cross-field correlation. Use nested when you must match multiple fields within the same array element. Defaulting to nested for everything wastes resources.

You learned how Elasticsearch indexes JSON objects:

• The default object type flattens fields and pools array values.
• The nested type indexes each array element separately to preserve relationships.
• Query nested fields with the nested query plus a path, and use inner_hits to find matching elements.
• Nested types cost more storage and require full re-indexing on element updates.