java-rust-dubbo

java-rust-dubbo is a small Dubbo consumer library for the Java/Rust REST framework.

Use it when your REST application is written in Java, your HTTP server is handled by the Rust native layer, and you only need to call Dubbo providers from your Java services without bringing a full Spring Boot or full Apache Dubbo runtime into the REST process.

The library keeps the programming model simple:

Your handlers, services, and components stay in Java.
The HTTP I/O plane stays in Rust Hyper through the Java/Rust REST framework.
Dubbo calls can use a Rust native transport for lower JVM RSS.
ZooKeeper and the official Dubbo/Netty client stack are optional, not default requirements.

The current aligned release is java-rust-dubbo:0.5.0 with rust-java-rest:4.0.0. It adds build-time generated native clients, declarative provider bindings, and a reusable bounded read retry policy. Existing bounded native thread stacks, provider restart handling, and the exclusive blocking or tokio-demux transport planes remain in place.

When To Use It

Recommended for:

REST services that need to call classic Dubbo providers.
Low-RSS Java services where loading the full Dubbo client stack is too expensive.
Kubernetes deployments where provider addresses can come from service DNS, config, or a sidecar-generated provider list.
Read-heavy or JSON-returning Dubbo methods where the provider can return a byte[] JSON body.
Java/Rust REST framework applications that want explicit setup instead of auto-configuration magic.

Use the official Dubbo stack instead when you need full Dubbo governance, config-center, metadata-center, official routers, callbacks, generic invocation, Triple, REST protocol, or full filter-chain compatibility.

Maven

<dependency>
  <groupId>com.reactor</groupId>
  <artifactId>java-rust-dubbo</artifactId>
  <version>0.5.0</version>
</dependency>

This artifact is published to GitHub Packages, not Maven Central. If the repository or package is private, the consuming project must also declare the GitHub Packages Maven repository and authenticate with a GitHub personal access token.

Add this repository to the consuming project's pom.xml:

<repositories>
  <repository>
    <id>github</id>
    <name>GitHub Packages</name>
    <url>https://maven.pkg.github.com/esasmer-dou/java-rust-dubbo</url>
  </repository>
</repositories>

Then add credentials to the consumer machine's ~/.m2/settings.xml. The <id> must match the repository id above:

<settings xmlns="http://maven.apache.org/SETTINGS/1.0.0"
          xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
          xsi:schemaLocation="http://maven.apache.org/SETTINGS/1.0.0 https://maven.apache.org/xsd/settings-1.0.0.xsd">
  <servers>
    <server>
      <id>github</id>
      <username>YOUR_GITHUB_USERNAME</username>
      <password>${env.GITHUB_PACKAGES_TOKEN}</password>
    </server>
  </servers>
</settings>

The token must be a classic GitHub PAT with read:packages. For private repositories, also grant repository access to the private repo, commonly through the repo scope. If your organization uses SSO, authorize the token for the organization. Do not put the token directly in pom.xml.

If a collaborator can read the repository but still receives 401 or 404 from Maven, check the package settings on GitHub. The package should either inherit access from esasmer-dou/java-rust-dubbo, or the collaborator/team should be granted at least Read access to the package.

In native mode this dependency is intentionally small. It does not pull ZooKeeper, Netty, Hessian Lite, or the official Dubbo client stack into your application unless you choose to add and use the official mode.

For the smallest static-provider native setup, use the native-static classifier instead of the full JAR:

<dependency>
  <groupId>com.reactor</groupId>
  <artifactId>java-rust-dubbo</artifactId>
  <version>0.5.0</version>
  <classifier>native-static</classifier>
</dependency>

Use this classifier only when all of these are true:

You use reactor.dubbo.transport=native.
You set reactor.dubbo.providers=host:port,....
Your hot Dubbo method is the low-overhead no-argument byte[] path.
You do not need Java ZooKeeper discovery, official Dubbo mode, or Java Hessian argument encode/decode in the consumer.

If you need ZooKeeper discovery, argument-bearing Dubbo methods, DTO decoding, official Dubbo compatibility, or full governance behavior, use the normal dependency without the classifier. The normal artifact keeps those compatibility paths available.

The Java/Rust framework native library must also be present. In rust-java-rest, the framework loads that native library for you. In standalone tests, make sure rust_hyper is available through java.library.path.

Native Dubbo transport requires Dubbo native ABI 7. The aligned rust-java-rest:4.0.0 runtime reports REST ABI 24, Dubbo ABI 7, and Redis ABI 6. Framework startup verifies the packaged source revision and platform hash; NativeDubboBridge also checks the Dubbo ABI before the first native client is created. Do not copy a DLL/SO from an older framework release into a newer image.

Public API Boundary

Use the public classes under com.reactor.rust.dubbo, com.reactor.rust.dubbo.config, com.reactor.rust.dubbo.support, and com.reactor.rust.dubbo.provider in application code.

Common consumer classes:

DubboConsumerConfig
DubboReferenceSpec
NativeDubboConsumerClient
NativeDubboConsumers
NativeDubboMethodInvoker
DubboConsumerSupport
com.reactor.rust.dubbo.codegen.GenerateNativeDubboClient from the build-only codegen classifier

Common provider classes:

DubboApplicationProperties
DubboProviderApplication
DubboProviderRuntimeTuning
DubboProviderSupport
DubboServiceBinding
PlainDubboProvider
ZookeeperDubboProviderRegistration
com.reactor.rust.dubbo.provider.jdbc.JdbcRepository
com.reactor.rust.dubbo.provider.jdbc.HikariDataSources

Classes under com.reactor.rust.dubbo.internal.* are implementation details. They are separated by responsibility so the runtime is easier to maintain:

internal.nativeclient: native Dubbo transport reference, codec, descriptor, and native provider watcher.
internal.direct: optional official Dubbo direct invoker path.
internal.registry: ZooKeeper provider discovery and Dubbo URL helpers.
internal.runtime: Dubbo runtime model and low-RSS Netty tuning.
internal.util: small runtime utilities.

Do not import internal.* from your service. Those packages can change between minor or patch releases without source compatibility guarantees. Source compatibility is guaranteed only for the public API listed above.

Declarative Consumer And Provider Helpers

DubboConsumerSupport and DubboProviderApplication remove repeated configuration and lifecycle code. They do not discover business services automatically. Configuration selects the active runtime surface, while your code still lists every provider interface and consumer adapter explicitly.

Recommended consumer example:

@GenerateNativeDubboClient(
        service = CatalogService.class,
        generatedName = "CatalogClient",
        retryReads = true,
        retryProperty = "catalog.read-retry-on-io-error",
        exposeMetrics = true)
final class CatalogClientDefinition {
    private CatalogClientDefinition() {}
}

The annotation processor generates CatalogClient.create(client, support) and one exact async method per Dubbo interface method. A byte[] return also gets a ...NativeJsonAsync() method that can keep the response body in Rust native memory. Method descriptors, return types, group, version, retry policy, and optional metrics access are fixed at build time. The generated wrapper does not use the dynamic proxy path. NativeDubboConsumerClient resolves each Java interface Method once while the client is created; request handling does not use proxy dispatch, method-name lookup, or reflection.

Provider example:

public final class CatalogProviderApplication {
    public static void main(String[] args) throws Exception {
        DubboProviderApplication.run(
                "provider.properties",
                "catalog-provider",
                CatalogProviderModule.INSTANCE);
    }
}

The named module contains only the explicit resource and service plan:

public void configure(DubboProviderApplication.ModuleContext context) {
    CatalogRepository repository = context.manage(
            CatalogRepository.fromProperties(context.properties()));
    context.services(
            service(CatalogService.class, new CatalogServiceImpl(repository)),
            service(CustomerQueryService.class, new CustomerQueryServiceImpl(repository)));
}

The application declares the service and resource. The library owns export, registry registration, low-RSS provider defaults, startup rollback, shutdown hooks, and reverse-order resource cleanup. The builder remains available for unusual embedded lifecycle requirements.

Declarative providers also get a bounded Dubbo/Netty executor. These properties are mapped to the exported Dubbo URL before the server starts:

dubbo.provider.executor.thread-pool=eager
dubbo.provider.executor.core-threads=1
dubbo.provider.executor.max-threads=8
dubbo.provider.executor.queue-capacity=16
dubbo.provider.executor.idle-timeout-ms=30000
dubbo.provider.executor.io-threads=1

This avoids Dubbo's default 200 handler-thread ceiling in small pods. cached is rejected because it is not a bounded production choice. Direct callers of the six-argument ProviderConfig constructor retain Dubbo defaults; DubboProviderSupport and DubboProviderApplication apply the bounded configuration above. Override it only with thread, RSS, p99, and rejection evidence.

BEST: keep the service list explicit and move only duplicated lifecycle code to these helpers. ANTI-PATTERN: adding an automatic provider scanner that exports every interface on the classpath.

DubboServiceBinding.service(...) is startup metadata only. It validates that the implementation matches the interface and registers the explicit list before Dubbo starts. It does not add a wrapper, lookup, allocation, or dispatch layer to a provider invocation.

Provider-side DB helpers are optional. Use them when a plain Dubbo provider needs the same low-boilerplate JDBC/Hikari lifecycle pattern as the sample provider. They do not generate SQL and they do not map rows automatically. Your provider still owns SQL, indexes, row mapping, transaction boundaries, and write idempotency. Hikari pool tuning keys have small built-in defaults, so a provider can start with only JDBC URL, driver, username and password.

Quick Start

This section shows the complete flow for adding a Dubbo consumer to a Java/Rust REST application.

1. Add The Dependency

Add the Maven dependency shown above to your application.

If your application pulls from the private GitHub package, also add the GitHub Packages repository and ~/.m2/settings.xml credentials from the Maven section. Adding only the dependency is not enough because Maven Central does not host this artifact.

If your project is already using the Java/Rust REST framework, the native library is usually loaded by the framework. You do not need a Dubbo Spring Boot starter.

2. Add Basic Properties

For the lowest-RSS setup, start with native transport and static providers:

reactor.runtime.profile=micro-dubbo
reactor.dubbo.enabled=true
reactor.dubbo.transport=native
reactor.dubbo.runtime-profile=micro-dubbo
reactor.dubbo.providers=catalog-provider:20880
reactor.dubbo.timeout-ms=800
reactor.dubbo.retries=0
reactor.dubbo.max-inflight=32
reactor.dubbo.max-response-bytes=8388608
reactor.dubbo.native-connections-per-endpoint=1
reactor.dubbo.native-max-idle-connections-per-endpoint=1
reactor.dubbo.native-idle-connection-ttl-ms=30000
reactor.dubbo.native-async-workers=1
reactor.dubbo.native-async-queue-capacity=32

What this means:

transport=native tells the library to use the Rust Dubbo data-plane.
runtime-profile=micro-dubbo keeps native workers, queues, refresh queues, and Netty fallback settings narrow.
providers=host:port tells NativeDubboConsumerClient where the provider is. With this set, the native low-RSS path does not start Java ZooKeeper.
retries=0 keeps latency predictable and avoids duplicate provider calls.
max-inflight and native queue settings prevent unlimited memory growth under load.

If you leave reactor.dubbo.providers empty, NativeDubboConsumerClient uses ZooKeeper discovery. That is supported, but it adds Java ZooKeeper classes and at least a small discovery thread footprint. Use it when provider failover must come from ZooKeeper; otherwise prefer static providers, Kubernetes service DNS, or a sidecar-generated provider list.

For very small pods, also tune the host JVM. The library can keep Dubbo small, but OpenJ9 still needs an explicit small CPU view to avoid extra JVM worker/JIT footprint:

-Xms8m -Xmx48m -Xss256k -Xquickstart -Xtune:virtualized -Xshareclasses:none -XX:ActiveProcessorCount=1

For very low traffic services only, adding -Xnojit can reduce RSS further. It is not a general throughput default; measure RPC p99 and CPU before using it in production.

3. Define The Dubbo Service Interface

Use the same interface name and method signature that the provider exports.

For the fastest current native path, a no-argument method returning byte[] is preferred when the provider can return ready-to-send JSON:

package com.example.catalog;

public interface CatalogProviderApi {
    byte[] nestedCatalogJson();
}

The returned byte[] can be JSON, MessagePack, protobuf bytes, or any binary payload your HTTP layer knows how to represent.

For command/read-with-parameter routes where the request body is already JSON, the native path also supports these low-overhead signatures:

public interface CustomerCommandApi {
    byte[] createCustomer(byte[] commandJson);

    byte[] patchCustomer(long customerId, byte[] commandJson);
}

These signatures keep request encoding in the Rust Hessian subset and keep the provider JSON response as a native HTTP body handle. Use typed record/list methods only when the consumer must inspect domain fields.

For REST JSON pass-through, prefer the native HTTP response handle path when possible:

public CompletableFuture<ResponseEntity<RawResponse>> catalog() {
    return invoker.invokeNativeJsonResponseAsync()
            .thenApply(handle -> ResponseEntity.ok(RawResponse.nativeResponse(handle.nativeId())));
}

This keeps the provider JSON body in Rust native memory and sends only a small response id through Java. Use the older invokeAsync().thenApply(bytes -> RawResponse.json(bytes)) path when the Java handler must inspect, transform, validate, or log the response bytes.

4. Generate The Native Client At Build Time (Recommended)

Add the small build-only classifier to the consuming application. provided keeps it out of the runtime dependency surface:

<dependency>
  <groupId>com.reactor</groupId>
  <artifactId>java-rust-dubbo</artifactId>
  <version>0.5.0</version>
  <classifier>codegen</classifier>
  <scope>provided</scope>
</dependency>

Register the processor in maven-compiler-plugin:

<annotationProcessorPaths>
  <path>
    <groupId>com.reactor</groupId>
    <artifactId>java-rust-dubbo</artifactId>
    <version>0.5.0</version>
    <classifier>codegen</classifier>
  </path>
</annotationProcessorPaths>
<annotationProcessors>
  <annotationProcessor>com.reactor.rust.dubbo.codegen.NativeDubboClientProcessor</annotationProcessor>
</annotationProcessors>

Declare the contract once:

@GenerateNativeDubboClient(
        service = CatalogProviderApi.class,
        generatedName = "CatalogClient",
        retryReads = true,
        retryProperty = "catalog.read-retry-on-io-error",
        exposeMetrics = true,
        group = "catalog",
        version = "1.0.0")
final class CatalogClientDefinition {
    private CatalogClientDefinition() {}
}

Create the generated client once during startup:

DubboConsumerSupport support = DubboConsumerSupport
        .fromProperties(PropertiesLoader.getAll());
NativeDubboConsumerClient transport = NativeDubboConsumers.create(support.config());
CatalogClient catalog = CatalogClient.create(transport, support);

The generated class exposes exact methods such as nestedCatalogJsonAsync() and, for a byte[] result, nestedCatalogJsonNativeJsonAsync(). The latter returns NativeResponseHandle and avoids materializing the provider JSON body as a Java byte[] when the REST endpoint only passes it through.

BEST: use generation for stable Dubbo interfaces. It removes handwritten descriptor and retry boilerplate without adding runtime magic. ACCEPTABLE: use the manual API below for dynamic or unusual embedded integrations. ANTI-PATTERN: create a reflective proxy or resolve method descriptors on every request.

The generator includes methods inherited from parent interfaces. Contract validation also runs at build time. Overloaded methods, void methods, generic service interfaces, and unresolved generic method types are rejected with a compiler error. Keep the Dubbo transport contract explicit and non-generic; use concrete records, primitives, strings, collections, or byte[] payloads.

5. Manual Client Wiring (Supported Alternative)

Create the consumer once and reuse it. Do not create a Dubbo client per request.

import com.example.catalog.CatalogProviderApi;
import com.reactor.rust.di.annotation.Bean;
import com.reactor.rust.di.annotation.Configuration;
import com.reactor.rust.di.annotation.PreDestroy;
import com.reactor.rust.dubbo.DubboReferenceSpec;
import com.reactor.rust.dubbo.NativeDubboConsumerClient;
import com.reactor.rust.dubbo.NativeDubboConsumers;
import com.reactor.rust.dubbo.NativeDubboMethodInvoker;
import com.reactor.rust.dubbo.NativeResponseHandle;

@Configuration
public final class DubboClientConfiguration {
    private final NativeDubboConsumerClient client = NativeDubboConsumers.create();

    @Bean
    public CatalogClient catalogClient() {
        DubboReferenceSpec<CatalogProviderApi> spec = DubboReferenceSpec
                .builder(CatalogProviderApi.class)
                .timeoutMs(1000)
                .retries(0)
                .check(false)
                .lazy(true)
                .build();

        NativeDubboMethodInvoker<byte[]> invoker =
                client.method(spec, "nestedCatalogJson", byte[].class);

        return new CatalogClient(invoker);
    }

    @PreDestroy
    public void shutdown() {
        client.close();
    }
}

Important points:

The NativeDubboConsumerClient is application-scoped.
The method invoker is created once and reused.
client.close() releases native clients, connections, and watcher resources.

6. Wrap The Invoker In A Java Client

This wrapper keeps the rest of your application clean. Your handlers and services do not need to know about Dubbo internals.

import com.reactor.rust.dubbo.NativeDubboMethodInvoker;
import com.reactor.rust.dubbo.NativeResponseHandle;

import java.util.concurrent.CompletableFuture;

public final class CatalogClient {
    private final NativeDubboMethodInvoker<byte[]> nestedCatalogJson;

    public CatalogClient(NativeDubboMethodInvoker<byte[]> nestedCatalogJson) {
        this.nestedCatalogJson = nestedCatalogJson;
    }

    public byte[] nestedCatalogJson() {
        return nestedCatalogJson.invoke();
    }

    public CompletableFuture<byte[]> nestedCatalogJsonAsync() {
        return nestedCatalogJson.invokeAsync();
    }

    public CompletableFuture<NativeResponseHandle> nestedCatalogNativeJsonAsync() {
        return nestedCatalogJson.invokeNativeJsonResponseAsync();
    }
}

Use invoke() for simple synchronous service code.

Use invokeAsync() when your REST framework handler can return CompletionStage. That keeps request workers from waiting on the Dubbo response.

7. Use It From A Handler Or Service

Example with a raw JSON response:

import com.reactor.rust.annotations.RustRoute;
import com.reactor.rust.di.annotation.Autowired;
import com.reactor.rust.http.RawResponse;
import com.reactor.rust.http.ResponseEntity;

import java.util.concurrent.CompletableFuture;

public final class CatalogHandler {
    @Autowired
    private CatalogClient catalogClient;

    @RustRoute(
            method = "GET",
            path = "/api/v1/catalog",
            responseType = RawResponse.class
    )
    public CompletableFuture<ResponseEntity<RawResponse>> catalog() {
        return catalogClient.nestedCatalogNativeJsonAsync()
                .thenApply(handle -> ResponseEntity.ok(RawResponse.nativeResponse(handle.nativeId())));
    }
}

If the provider is down or overloaded, handle the failed future and return a controlled HTTP response from your application layer.

7. Add A Route-Level Limit

Native limits protect the Dubbo client. Your HTTP route should also have its own concurrency policy.

Use a small route-level bulkhead when low RSS matters. Use a larger but still bounded limit for balanced throughput. Avoid unbounded queues.

8. Verify Before Production

Before shipping:

Start the app with Dubbo disabled and measure RSS.
Start the app with Dubbo enabled and measure RSS.
Call the Dubbo endpoint under c64, c256, c512, and c1000 load.
Restart the provider and confirm reconnect behavior.
Make the provider slow and confirm timeout behavior.
Confirm shutdown releases the native client.

Configuration Reference

All properties start with reactor.dubbo..

Values can come from a supplied Properties object, Java system properties, or environment variables. System properties and environment variables override the supplied Properties.

For environment variables, use uppercase and replace . / - with _.

Example:

REACTOR_DUBBO_TIMEOUT_MS=1000
REACTOR_DUBBO_NATIVE_ASYNC_WORKERS=8

Basic Properties

Property	Default	What It Does	When To Change It
`reactor.dubbo.application-name`	`rust-java-rest-dubbo-consumer`	Name used by the consumer. Mostly useful for diagnostics and official-mode metadata.	Change it to your service name, for example `orders-api`.
`reactor.dubbo.transport`	`native`	Chooses the transport. `native` uses Rust; `official` uses the optional Apache Dubbo stack.	Keep `native` for low RSS. Use `official` only when you need full Dubbo behavior.
`reactor.dubbo.providers`	empty	Static provider list in `host:port,host2:port` format. When set, `NativeDubboConsumerClient` skips Java ZooKeeper discovery.	Set this in Kubernetes/service-DNS/sidecar mode. This is the simplest low-RSS path.
`reactor.dubbo.registry-address`	`zookeeper://127.0.0.1:2181`	ZooKeeper address used only when `providers` is empty.	Change it only if you want ZooKeeper discovery.
`reactor.dubbo.registry-root`	`dubbo`	Root path for provider nodes in ZooKeeper.	Change it if your Dubbo providers are registered under a custom root.
`reactor.dubbo.registry-check`	`false`	Controls whether registry availability should be treated as startup-critical in discovery mode.	Keep `false` for rolling deployments. Use readiness checks if startup must fail when registry is missing.

Call Behavior

Property	Default	What It Does	When To Change It
`reactor.dubbo.timeout-ms`	`1000`	Per-call timeout. It bounds how long the consumer waits for a provider response.	Lower it for strict p99. Raise it only if the provider normally needs more time.
`reactor.dubbo.retries`	`0`	Number of extra attempts after a failed call.	Keep `0` unless the provider method is idempotent. Retries can multiply load and increase p99.
`reactor.dubbo.check`	`false`	Controls whether a reference should require provider availability during startup.	Keep `false` for rolling provider restarts. Add application readiness checks if the dependency is mandatory.
`reactor.dubbo.lazy`	`false`	Indicates lazy reference behavior for official-style compatibility. Native static mode still creates the native client when the reference is used.	Use only if you need delayed reference behavior.
`reactor.dubbo.protocol`	`dubbo`	Protocol name expected by classic Dubbo providers.	Usually do not change.
`reactor.dubbo.serialization`	`hessian2`	Serialization name used by classic Dubbo.	Keep `hessian2` for current native mode.
`reactor.dubbo.cluster`	`failfast`	Supported values are `failfast` and `failover`.	Use `failfast` for predictable latency. Use `failover` only for idempotent calls.
`reactor.dubbo.loadbalance`	`random`	Supported values are `random` and `roundrobin`.	Use `random` by default. Use `roundrobin` when you want simple even distribution.

Native Resource Limits

Property	Default	What It Does	When To Change It
`reactor.dubbo.max-inflight`	`32`	Max concurrent native Dubbo calls per reference. This is a bulkhead.	Lower it for low RSS and fail-fast behavior. Raise it only with enough provider capacity.
`reactor.dubbo.max-response-bytes`	`8388608`	Max Dubbo response frame size accepted by native mode.	Raise it only if your provider really returns larger payloads. Prefer streaming or smaller responses when possible.
`reactor.dubbo.native-connections-per-endpoint`	`1`	Max keepalive TCP connections per provider endpoint. One connection carries one in-flight call at a time.	Use `2` for a measured two-lane DB write path. Balanced throughput may use more only after a provider-capacity gate.
`reactor.dubbo.native-max-idle-connections-per-endpoint`	`1`	Max completed connections retained for reuse per endpoint.	Keep it at or below `native-connections-per-endpoint`. Lower it when idle RSS matters.
`reactor.dubbo.native-idle-connection-ttl-ms`	`30000`	Drops an idle pooled connection before reuse after this age. Older idle sockets are reconnected before request bytes are sent.	Keep `30000` unless the provider closes idle connections sooner. Then choose a smaller value than the provider idle timeout. Valid range: `1000..3600000`.
`reactor.dubbo.native-async-workers`	`1`	Native worker count used for async Dubbo calls.	Low RSS can keep this small. Raise it with the connection count only after load tests.
`reactor.dubbo.native-async-queue-capacity`	`32`	Bounded queue for native async calls. If full, calls fail fast instead of growing memory.	Keep bounded. Raise together with workers and route-level limits.
`reactor.dubbo.native-async-transport`	`blocking`	Async execution model. `blocking` uses the smallest worker model; `tokio-demux` uses Rust async request-id demux over provider connections.	Use `blocking` for lowest RSS and low traffic. Use `tokio-demux` for read-heavy/high-concurrency routes after a Docker RSS + p99 gate.
`reactor.dubbo.native-thread-stack-bytes`	`262144`	Stack size for native Dubbo worker and Tokio threads. Valid range: `131072..8388608`.	Keep `262144` for memory-first profiles. Raise only after a stack-overflow proof; lowering requires full route smoke tests.

Transport selection is exclusive per native client. blocking allocates blocking endpoint pools and no async endpoint pools. tokio-demux allocates async endpoint pools and no blocking endpoint pools; the synchronous Java facade bridges to that same Tokio runtime. Verify this with nativeDubboBlockingEndpointPools and nativeDubboAsyncEndpointPools metrics. Do not enable both resource models to hide overload.

Blocking transport validates a connection that has been idle for at least 100 ms before reuse. A closed provider socket is discarded before the request starts. The client never blindly retries a write after request bytes may have been sent. Watch nativeDubboIdleConnectionsExpired, nativeDubboIdleConnectionValidations, nativeDubboStaleIdleConnectionsDiscarded, and nativeDubboStaleConnectionRetries when testing provider restarts.

ZooKeeper And Official-Mode Properties

Property	Default	What It Does	When To Change It
`reactor.dubbo.registry-timeout-ms`	`3000`	Timeout for ZooKeeper connection/operations.	Change only in ZooKeeper discovery mode.
`reactor.dubbo.registry-session-timeout-ms`	`30000`	ZooKeeper session timeout.	Change only if your registry environment requires it.
`reactor.dubbo.connections`	`1`	Official-mode connection setting. Native mode uses `native-connections-per-endpoint` instead.	Mostly relevant for official mode.
`reactor.dubbo.share-connections`	`1`	Official-mode shared connection setting.	Mostly relevant for official mode.
`reactor.dubbo.refer-thread-num`	`1`	Worker count for provider refresh in ZooKeeper mode. It is not a request worker pool.	Keep small unless you have many references refreshing at once.
`reactor.dubbo.runtime-profile`	`micro-dubbo`	Descriptive profile value. Host applications may map it to runtime presets.	Use `micro-dubbo` or `balanced-dubbo` according to your measured deployment profile.

Suggested Starting Profiles

Micro Dubbo / lowest RSS:

reactor.dubbo.runtime-profile=micro-dubbo
reactor.dubbo.transport=native
reactor.dubbo.providers=provider-1:20880
reactor.dubbo.retries=0
reactor.dubbo.timeout-ms=800
reactor.dubbo.max-inflight=32
reactor.dubbo.native-connections-per-endpoint=1
reactor.dubbo.native-idle-connection-ttl-ms=30000
reactor.dubbo.native-async-workers=1
reactor.dubbo.native-async-queue-capacity=32

Balanced:

reactor.dubbo.runtime-profile=balanced-dubbo
reactor.dubbo.transport=native
reactor.dubbo.providers=provider-1:20880,provider-2:20880
reactor.dubbo.retries=0
reactor.dubbo.timeout-ms=1200
reactor.dubbo.max-inflight=512
reactor.dubbo.native-connections-per-endpoint=16
reactor.dubbo.native-async-workers=8
reactor.dubbo.native-async-queue-capacity=1024

Use micro Dubbo when memory is the first priority and overload can return controlled 503 responses.

Use balanced when Dubbo throughput matters more and you have enough provider capacity.

Native Mode vs Official Mode

Native mode is the default because it keeps the hot JVM smaller. It avoids loading the official Dubbo ReferenceConfig, registry directory, official remoting, Netty, Java ZooKeeper client, Hessian Lite on the hot no-arg byte[] path, config-center, metadata-center, tracing, metrics modules, and extra transport codecs.

Official mode is still useful when correctness depends on full Dubbo behavior. Use DubboConsumers/DubboConsumerClient only for this compatibility path. If you use official mode, add the required optional dependencies explicitly and expect higher RSS.

Current Native-Mode Limits

Native mode is intentionally focused. It does not currently implement:

Dubbo config-center and metadata-center.
Consumer registration under /consumers.
Official Dubbo routers and governance rules.
Triple, REST protocol, HTTP/2, callbacks, generic invocation, or full official filter-chain behavior.
Multiplexed request demux over a single Dubbo connection.

For many REST-to-Dubbo consumer cases this is enough. If your system depends on the missing features, use official mode or put a full Dubbo integration service behind a smaller REST process.

Build

mvn clean verify

Release artifacts are produced under target/:

java-rust-dubbo-0.5.0.jar
java-rust-dubbo-0.5.0-native-static.jar
java-rust-dubbo-0.5.0-codegen.jar (build time only)
java-rust-dubbo-0.5.0-sources.jar

Name		Name	Last commit message	Last commit date
Latest commit History 22 Commits
.github/workflows		.github/workflows
benchmark		benchmark
docs		docs
src		src
.gitignore		.gitignore
README.md		README.md
README.tr.md		README.tr.md
pom.xml		pom.xml

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

java-rust-dubbo

When To Use It

Maven

Public API Boundary

Declarative Consumer And Provider Helpers

Quick Start

1. Add The Dependency

2. Add Basic Properties

3. Define The Dubbo Service Interface

4. Generate The Native Client At Build Time (Recommended)

5. Manual Client Wiring (Supported Alternative)

6. Wrap The Invoker In A Java Client

7. Use It From A Handler Or Service

7. Add A Route-Level Limit

8. Verify Before Production

Configuration Reference

Basic Properties

Call Behavior

Native Resource Limits

ZooKeeper And Official-Mode Properties

Suggested Starting Profiles

Native Mode vs Official Mode

Current Native-Mode Limits

Build

Documentation

About

Uh oh!

Releases 12

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

java-rust-dubbo

When To Use It

Maven

Public API Boundary

Declarative Consumer And Provider Helpers

Quick Start

1. Add The Dependency

2. Add Basic Properties

3. Define The Dubbo Service Interface

4. Generate The Native Client At Build Time (Recommended)

5. Manual Client Wiring (Supported Alternative)

6. Wrap The Invoker In A Java Client

7. Use It From A Handler Or Service

7. Add A Route-Level Limit

8. Verify Before Production

Configuration Reference

Basic Properties

Call Behavior

Native Resource Limits

ZooKeeper And Official-Mode Properties

Suggested Starting Profiles

Native Mode vs Official Mode

Current Native-Mode Limits

Build

Documentation

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases 12

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages