Unified Search Index

The design principle of indexing records of many types into a single search index whose schema is shaped by user information needs, not by the source data model. The canonical statement: one field per use case of how people want to find things — not one field per database column or one index per table.

The Idea

When making a relational database searchable, the instinct is to mirror the database: an index per table, foreign keys preserved, relationships as nested objects. This produces fragmented, slow search. The unified-index approach instead:

1. One index holds every searchable record type.
2. A type field distinguishes records and drives faceting and filtering.
3. The schema is a small, fixed set of purpose fields (names, aka, notes, address, amount, dates, permissions, url…) that every record type maps onto.
4. Relationships are denormalized into flat multivalued fields at index time rather than joined at query time.

Why It Works

• A single query searches across all entity types at once — a user searching a brand finds its products, the brand page, and related entities together.
• Adding a new source table is a few lines of field mapping, not a new index and query path.
• Document-level concerns (access control via a permissions field, soft-delete via a deleted date) are handled uniformly for every type.
• Avoids nested fields, which are slower and more complex than flat multivalued fields.

Trade-offs

• Requires up-front analysis of information needs — what users actually search for — rather than mechanically reflecting the schema.
• Denormalization shifts cost to index time and write-side maintenance (resolved values must be re-indexed when the source changes).

Related Concepts

• Federated vs Unified Search — the topic hub comparing index-time consolidation against the query-time alternative, Federated Search
• Denormalization for Search — the mechanism that makes a unified schema possible
• Multi-Tenancy in Search — the same one-index-for-many question when the “many” are customers rather than record types
• Full-Text Search
• Faceted Search — the type field is the natural primary facet
• Search Architecture

Articles

• How to Really Design Search for a Database — Bonsai / Max Irwin; the 14-field universal schema

Case Studies

• Bonsai - Designing Search for a Relational Database
• Netflix - Content Search Architecture — federated entities indexed into a searchable whole