What’s New in Dubbo 3.0 Preview? Async Filters, Reactive Support & Cloud‑Native Insights

Dubbo, an open‑source Java RPC framework, has been widely adopted since its 2011 release. Starting from December last year, Dubbo 3.0 entered development and attracted community attention.

Key questions

1. Why was the first open‑source version 2.0? Dubbo actually had a 1.x series with a different protocol and model.

2. Is Alibaba still using the open‑source version? Yes, Alibaba uses the open‑source Dubbo in many production systems, while its core e‑commerce department still runs HSF 2.2, which is compatible with Dubbo.

Dubbo 3.0 Preview Highlights

Dubbo 3.0 adds four major improvements:

Asynchronous Filter chain – unlike the async support in 2.7, 3.0 introduces true end‑to‑end async at the protocol level, enabling streaming, back‑pressure and rate‑limiting.

Reactive programming support – implements Reactive Streams interfaces so users can benefit from reactive semantics such as event‑driven integration, smarter load‑balancing and adaptive fault‑tolerance.

Cloud‑native / Service‑Mesh exploration – integration with Envoy and future mesh capabilities, though full production readiness is still pending.

Fusion with Alibaba internal systems – the core will eventually be deployed in Alibaba’s production environment, ensuring performance, stability and governance.

1. Asynchronous Filter Chain Design

Dubbo’s extensible Filter SPI allows users to enhance functionality. The original invoke method returns a Result synchronously. In 3.0 the design is refactored to use CompletableFuture internally, providing a fully asynchronous call chain while keeping the external API compatible.

Key points of the refactor:

Framework internally uses a full‑async model; external code can still use synchronous calls.

Built‑in Filter chain supports asynchronous callbacks.

Typical workflow:

Consumer invokes a proxy, request passes through the Filter chain.

Invoker forwards the request asynchronously and obtains a Future.

The Future is wrapped into a Result object.

Each Filter may register a Listener to process the result when it arrives.

Proxy returns either the actual object (synchronous) or the Future (asynchronous) to the caller.

Result can be implemented either by inheriting CompletionStage or by composition with a Future, and new async methods are added to the Result interface.

2. Reactive Programming Support

Dubbo 3.0 adds Reactive Streams support via RSocket, allowing Mono/Flux types in service interfaces. The protocol encodes requests into RSocket Payloads and uses requestStream to transmit data. On the server side, the Payload is decoded, the business method returns a Flux, which is then mapped back to Payload for transmission.

Sample interface and implementation snippets illustrate how to define a service returning Mono<String> or Flux<String> and configure the protocol name as rsocket on both provider and consumer.

3. Async Generic Invocation

Dubbo’s generic API now includes async methods, enabling callers to obtain a CompletableFuture directly for generic services.

4. Performance Considerations

While internal async does not speed up a single call, it reduces thread‑blocking, improves overall concurrency and throughput, which is beneficial for latency‑sensitive RPC workloads.

For more details and sample projects, see the Dubbo 3.0 preview repository on GitHub.