Nereus Delay documentation
Nereus Delay is a Java 21 library and service core for durable delayed-message scheduling across Kafka and Pulsar destinations. V1 is complete: the model and implementation separate durable command ordering, Delay Shard application, destination scheduling, producer evidence, and recovery.
The V1 semantic and protocol baseline is frozen as V1-FROZEN-2026-08-13. These pages explain how to choose an entry path, understand the lifecycle, integrate Kafka or Pulsar, operate the service, and reason about failure boundaries.
Start here
- Why Nereus Delay? — goals, boundaries, and why scheduling needs durable state.
- Architecture — the end-to-end components and their authority boundaries.
- Choose an entry path — Direct SDK, Delay Gateway,
MANAGED, andAUTO_FAST. - Lifecycle walkthrough — one message from preparation through outcome and recovery.
- Delivery time and action time — the not-before timing model.
- Delay Shard and materialized state — how the log, RocksDB, and recovery unit fit together.
- Kafka and Pulsar — source and destination semantics by Broker.
- Correctness invariants — the durable ordering rules behind failure recovery.
- Project status — V1 completion evidence and the boundary for subsequent evolution.