ADR Component Demo

Use PostgreSQL for persistence

January 15, 2026

We need a relational database that supports complex queries and full-text search. The team has strong SQL experience and needs ACID guarantees for financial transactions.

Use PostgreSQL 16 as the primary data store with pg_trgm for full-text search.

• Good: Mature ecosystem with excellent tooling
• Good: Strong ACID compliance for financial data
• Bad: Vertical scaling limits at high write volumes
• Bad: Full-text search less capable than dedicated engines

Add Elasticsearch for search

February 20, 2026

PostgreSQL pg_trgm full-text search does not meet latency requirements for the product catalog search feature. Users expect sub-100ms results with typo tolerance and faceted filtering.

Add Elasticsearch 8 as a read-optimized search index alongside PostgreSQL. PostgreSQL remains the system of record; Elasticsearch is populated via CDC (Change Data Capture).

• Good: Sub-50ms search with typo tolerance and facets
• Good: PostgreSQL remains source of truth for writes
• Bad: Added operational complexity (two data stores)
• Bad: CDC pipeline introduces eventual consistency window

Adopt event sourcing for audit trail

March 10, 2026

Regulatory requirements mandate a complete audit trail of all financial transactions. The current approach of database triggers is fragile and does not capture intent.

Implement event sourcing for the payments domain using Apache Kafka as the event store, with PostgreSQL projections for query models.

• Good: Complete audit trail with business intent captured
• Good: Enables temporal queries and replay for debugging
• Bad: Significant learning curve for the team
• Bad: Increased infrastructure costs for Kafka cluster