cursus is designed around several core principles:
Simplicity over feature completeness: Single-node architecture with minimal dependencies, avoiding distributed consensus complexity
Separation of concerns: Clear boundaries between connection handling, topic management, and disk persistence
Asynchronous writes, synchronous reads: Publishers don’t block on disk I/O; consumers read directly from segments
Batching and buffering: Multiple levels of buffering balance throughput and latency
Per-partition isolation: Each partition operates independently, enabling parallelism
Observable: Built-in Prometheus metrics, health checks, and structured logging
These design choices prioritize operational simplicity and performance for single-node deployments while maintaining the core guarantees of a message broker (durability, ordering within partitions, at-least-once delivery).
Future versions plan to add distributed coordination using etcd for multi-node deployments.