# Research & Roadmap
Based on a review of the Rust ES ecosystem (cqrs-es, disintegrate, esrs, eventually-rs, kameo_es/SierraDB) and mature frameworks in other languages (Axon, EventFlow, Marten).
---
## Prioritized Work Items
### Core Improvements
#### Async traits
All competing frameworks are async-first. The original traits (`Repository`, `Commit`, `Get`, etc.) are synchronous, which blocks direct integration with:
- Async databases (sqlx, sea-orm)
- Async message brokers (rdkafka, lapin)
- Async web frameworks (axum handlers)
Foundation status: stream-aware repository/read-model/snapshot traits now form the async-only persistence surface. See [Repository Boundary](repositories.md). The SQL backends implement those traits directly instead of wrapping database I/O behind synchronous repository traits.
### Later
- **Postgres backend** — proves the trait design, makes the library production-usable. Use sqlx. The event-record storage contract is outlined in [Postgres Event Store Contract](postgres-event-store.md).
- **API docs** — `cargo doc` with doc comments on all public traits/types.
- **Publish to crates.io** — after the above items stabilize.
- **Domain service** — TBD whether to keep. Not documenting further until decided.
---
## Ecosystem Comparison (Reference)
### Feature Matrix
| Aggregates | Yes (PORS + macros) | Yes (trait) | Decision pattern | Yes (trait) | Yes (DDD) |
| Event storage | In-memory | PG/MySQL/Dynamo | PG/In-memory | PG only | PG/In-memory |
| Projections/Read models | Yes (3 strategies) | Query trait | EventListener | EventHandler (2 types) | Projection trait |
| Snapshots | Yes (frequency-based) | Yes | Via StateQuery | Yes (Nth event) | No |
| Outbox pattern | Yes (first-class) | No | No | Implicit (TransactionalEventHandler) | No |
| Service bus (pub/sub + P2P) | Yes | No | No | No | No |
| Saga support | Via aggregates | Via handlers | Via EventListener | Via handlers | Via Subscription |
| Event upcasting | **No** | **Yes** | No | No | Partial |
| Async | Partial | Yes | Yes | Yes | Yes |
| Testing framework | N/A (simple code) | Given-When-Then | In-memory store | Manual | AggregateRootBuilder |
| Optimistic concurrency | Yes | Yes | Yes (+ validation queries) | Yes (+ pessimistic) | Yes |
| Pessimistic locking | Yes (QueuedRepo) | No | No | Yes (lock_and_load) | No |
| Event metadata | Yes | Yes | Yes | Yes | Yes |
| Derive macros | digest, aggregate, ReadModel | No | Yes (Event, StateQuery) | No | No |
| Observability | No | No | No | Yes (tracing) | No |
| Pluggable infra (all concerns) | **Yes (6 traits)** | Partial (storage) | Partial (storage) | No (PG only) | Partial (storage) |
### distributed differentiators (things others don't have)
- **Outbox pattern as first-class** with fan-out and point-to-point routing
- **Service bus** with both pub/sub and send/listen patterns
- **Read model transactional commits** with decision flowchart for choosing consistency strategy
- **6 pluggable infrastructure traits** (storage, messaging, read model store, snapshot store, outbox publishing, locking)
- **Guard conditions on events** (`when = !self.completed`)
- **PORS approach** — domain objects are plain structs, not framework types
### Key competitors at a glance
**cqrs-es** (~463 GitHub stars, most mature)
- Best testing framework, event upcasting, 3 storage backends
- Serverless-first design
- Docs: [doc.rust-cqrs.org](https://doc.rust-cqrs.org/)
**disintegrate** (innovative approach)
- Decision pattern instead of aggregates
- Validation queries for fine-grained optimistic concurrency
- Derive macros for events and state queries
- Docs: [docs.rs/disintegrate](https://docs.rs/disintegrate)
**esrs** (production-proven at Prima.it)
- Transactional vs eventually-consistent event handlers
- Built-in tracing/observability
- Pessimistic locking option
- Docs: [docs.rs/esrs](https://docs.rs/esrs)
**eventually-rs** (pre-1.0, DDD-pure)
- Clean DDD abstractions
- Subscription mechanism for reactive processing
- Limited documentation
- Docs: [docs.rs/eventually](https://docs.rs/eventually)
**kameo_es + SierraDB** (successor to thalo, newest)
- Actor-based (kameo) + distributed event store (SierraDB)
- Historical + live subscriptions with guaranteed delivery
- Very new, worth watching