Search Engine & Full-Text
Priority: P1 (OPERATIONAL)
Full-text search implementation with Elasticsearch and database patterns.
- Pattern: CQRS (Command Query Responsibility Segregation).
- Write: To Primary Database (Postgres/MySQL). Source of Truth.
- Read (Complex): To Search Engine (Elasticsearch, OpenSearch, MeiliSearch). Optimized for filtering, fuzzy search, and aggregation.
Synchronization (The Hard Part)
- Dual Write (Anti-Pattern):
await db.save(); await es.index();.- Why: Partial failures leave data inconsistent. Slows down HTTP response.
- Event-Driven (Recommended):
- Service writes to DB.
- Service emits
EntityUpdated. - Event Handler (Async) pushes to Queue (BullMQ).
- Worker indexes document to Search Engine with retries.
- CDC (Golden Standard): Change Data Capture (Debezium). Connects directly to DB transaction log. No app conceptual overhead, but higher ops complexity.
Organization
- Module: Encapsulate the client in a
SearchModule. - Abstraction: Create generic
SearchService<T>helpers.indexDocument(id, body)search(query, filters)
- Mapping: Use
class-transformerto map Entities to "Search Documents". Search docs should be flatter than Relational entities constraints.
Testing
- E2E: Do not mock the search engine in critical E2E flows.
- Docker: Spin up
elasticsearch:8container in the test harness to verify indexing works.
Anti-Patterns
- No dual writes to DB + ES: Use event-driven or CDC pattern; dual writes risk partial failure inconsistency.
- No Elasticsearch for structured queries: Use DB indexes for filtering; ES is for full-text and complex search.
- No ES mocks in E2E search tests: Spin up
elasticsearch:8container to verify indexing behavior accurately.