In enterprise B2B software where thousands of concurrent transactions occur every second, maintaining database latency in single-digit milliseconds is a vital requirement. Here we explore the architecture powering Digital HAT's mission-critical platforms using .NET 9, Entity Framework Core, and PostgreSQL.
1. Decoupling Workloads with CQRS & MediatR
Traditional CRUD patterns suffer from row locks when heavy analytics queries share the same domain models with rapid transaction mutations. We solve this by decoupling execution paths:
- Commands: Strictly validated, minimal payload mutations executed under isolated atomic transactions.
- Queries: Zero-overhead queries utilizing
AsNoTracking()and direct DTO projection projections to bypass the EF Core change tracking state machine entirely.
2. High-Performance PostgreSQL Connection Pooling
Re-negotiating TLS handshakes and TCP sockets introduces 30-50ms latency per request. Leveraging Npgsql 9 with connection multiplexing keeps pooled sockets warm, dropping database connection acquisition time below 0.5ms under sustained load.