protoc plugin for compiling gRPC RPC services as Jersey/REST endpoints. Uses the HttpRule annotations also used by the grpc-gateway project to drive resource generation.

• Example Usage
• Operation modes
  • HTTP and gRPC
• Working with HTTP headers
  • Streaming
  • Errors
  • Headers in the main gRPC Metadata (deprecated)
• Streaming RPCs
• Error handling
  • Error Translation
    • Retry-After
  • Streaming RPCs
  • Overriding error handling
• JSON Serialization
  • Overriding JSON formatting
• Releases
• Project status
• Build Process

grpc-jersey requires a minimum of Java 8 at this time.

Snapshot artifacts are available on the Sonatype snapshots repository, releases are available on Maven Central.

Example provided here uses the gradle-protobuf-plugin but an example using Maven can be found in examples.

ext {
    protobufVersion = "3.5.0"
    grpcVersion = "1.8.0"
    grpcJerseyVersion = "0.3.0"
}

protobuf {
    protoc {
        artifact = "com.google.protobuf:protoc:${protobufVersion}"
    }
    plugins {
        grpc {
            artifact = "io.grpc:protoc-gen-grpc-java:${grpcVersion}"
        }
        jersey {
            artifact = "com.xorlev.grpc-jersey:protoc-gen-jersey:${grpcJerseyVersion}"
        }
    }
    generateProtoTasks {
        all()*.plugins {
            grpc {}
            jersey {}
        }
    }
}

You'll also have to be sure to include the jersey-rpc-support package in your service:

compile "com.xorlev.grpc-jersey:jersey-rpc-support:${grpcJerseyVersion}"

Running ./gradlew build and a protobuf definition that looks roughly like the below

syntax = "proto3";

option java_package = "com.fullcontact.rpc.example";
import "google/api/annotations.proto";

service TestService {
    rpc TestMethod (TestRequest) returns (TestResponse) {
        option (google.api.http).get = "/users/{id}";
    }
    rpc TestMethod2 (TestRequest) returns (TestResponse) {
        option (google.api.http) = {
            post: "/users/",
            body: "*";
        };
    }
    rpc StreamMethod1 (TestRequest) returns (stream TestResponse) {
            option (google.api.http).get = "/stream/{s}";
    }
}
message TestRequest {
    string id = 1;
}
message TestResponse {
    string f1 = 1;
}

Would compile into a single Jersey resource with one GET handler and one POST handler.

Rules can also be defined in a .yml file.

http:
  rules:
  - selector: TestService.TestMethod4
    get: /users/{id}
  - selector: TestService.TestMethod5
    get: /yaml_users/{s=hello/**}/x/{uint3}/{nt.f1}/*/**/test
  - selector: TestService.TestMethod6
    post: /users/
    body: "*"
    additionalBindings:
      - post: /yaml_users_nested
        body: "nt"

Rules defined this way must correspond to methods in the .proto files, and will overwrite any http rules defined in the proto. The path to your .yml file should be passed in as an option:

generateProtoTasks {
    all()*.plugins {
        grpc {}
        jersey {
            option 'yaml=integration-test-base/src/test/proto/http_api_config.yml'
        }
    }
}

or

    <configuration>
      <pluginId>grpc-jersey</pluginId>
      <pluginArtifact>com.xorlev.grpc-jersey:protoc-gen-jersey:0.1.4:exe:${os.detected.classifier}</pluginArtifact>
      <pluginParameter>yaml=integration-test-base/src/test/proto/http_api_config.yml</pluginParameter>
    </configuration>

grpc-jersey can operate in two different modes: direct invocation on service ImplBase or proxy via a client Stub. There are advantages and disadvantages to both, however the primary benefit to the client stub proxy is that RPCs pass through the same ServerInterceptor stack. It's recommended that the client stub passed into the Jersey resource uses a InProcessTransport if living in the same JVM as the gRPC server. A normal grpc-netty channel can be used for a more traditional reverse proxy.

The proxy stub mode is the default as of 0.2.0.

You can toggle the "direct invocation" mode by passing an option to the grpc-jersey compiler:

generateProtoTasks {
    all()*.plugins {
        grpc {}
        jersey {
            option 'direct'
        }
    }
}

You can find a complete example of each in the integration-test-proxy and integration-test-serverstub projects.

If you plan to run "dual stack", that is, services serving traffic over both HTTP and RPC, you can configure your service to share resources and the same interceptor stack and avoid TCP/serialization overhead using a mixture of in-process transport. The example below uses Dropwizard, but should be adaptable to any Jersey glue you like.

// Shared executor for both services.
Executor executor = Executors.newFixedThreadPool(config.rpcThreads,
        new ThreadFactoryBuilder().setNameFormat("grpc-executor-%d").build());

// Service stack. This is where you define your interceptors.
ServerServiceDefinition serviceStack = ServerInterceptors.intercept(
        new EchoTestService(),
        new GrpcLoggingInterceptor(),
        new MyAuthenticationInterceptor() // your interceptor stack.
);

// External RPC service.
Server externalService = NettyServerBuilder
        .forPort(config.rpcPort)
        .executor(executor)
        .addService(serviceStack)
        .build()
        .start();

// In-memory RPC service for grpc-jersey. This avoids serialization/TCP overheads while still sharing the
// same executor and service stack.
Server internalService = InProcessServerBuilder
        .forName("TestService")
        .executor(executor)
        .addService(serviceStack)
        .build()
        .start();

// In-process stub used by the generated Jersey resource.
TestServiceGrpc.TestServiceStub stub =
        TestServiceGrpc.newStub(InProcessChannelBuilder
                .forName("TestService")
                .usePlaintext(true)
                .directExecutor()
                .build());

// Dropwizard-specific: register shutdown handlers and the Jersey resource.

environment.lifecycle().manage(new Managed() {
    @Override
    public void start() throws Exception {
    }

    @Override
    public void stop() throws Exception {
        externalService.shutdown();
        internalService.shutdown();
    }
});

environment.jersey().register(new TestServiceGrpcJerseyResource(stub));

At this time, only streaming from server to client is supported. Client to server streaming will also be supported in the future, allowing for limited bi-directional streaming. Due to the limitations of the HTTP/1.1 protocol, server streaming will only begin once the client stream terminates.

grpc-jersey streams messages as newline-delimited JSON. As an example:

> GET /stream/hello?int3=2 HTTP/1.1
> Host: localhost:8080
> User-Agent: curl/7.51.0
> Accept: */*
>
< HTTP/1.1 200 OK
< Vary: Accept
< Content-Type: application/json;charset=utf-8
< Vary: Accept-Encoding
< Transfer-Encoding: chunked
<
{"request":{"s":"hello","uint3":0,"uint6":"0","int3":2,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}
{"request":{"s":"hello","uint3":0,"uint6":"0","int3":2,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}

Each message is encoded into a single line.

By default, the handler returns application/json;charset=utf-8, however if provided an Accept header of text/event-stream, grpc-jersey will provide data in a format usable by EventSource (Server-Sent Events):

> GET /stream/hello?int3=2 HTTP/1.1
> Host: localhost:8080
> User-Agent: curl/7.51.0
> Accept: text/event-stream
>
< HTTP/1.1 200 OK
< Vary: Accept
< Content-Type: text/event-stream;charset=utf-8
< Vary: Accept-Encoding
< Transfer-Encoding: chunked
<
data: {"request":{"s":"hello","uint3":0,"uint6":"0","int3":2,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}

data: {"request":{"s":"hello","uint3":0,"uint6":"0","int3":2,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}

NOTE: This only works for uses using the "proxy" configuration. Direct invocation mode does not support HTTP header manipulation.

grpc-jersey allows you to use and manipulate HTTP headers from within your RPC handler. To do so, you'll need to install a server interceptor into your RPC stack. If you're using the recommended "dual-stack" configuration, you can modify it like so:

  // Service stack. This is where you define your interceptors.
  ServerServiceDefinition serviceStack = ServerInterceptors.intercept(
          GrpcJerseyPlatformInterceptors.intercept(new EchoTestService()),
          new GrpcLoggingInterceptor(),
          new MyAuthenticationInterceptor() // your interceptor stack.
  );

Preferably, you'll want to use the helper provided by GrpcJerseyPlatformInterceptors. Future platform interceptors will be added here automatically, allowing your code to remain the same and take advantage of new functionality. However, you can also use just the HttpHeaderInterceptor directly should you desire:

  // Service stack. This is where you define your interceptors.
  ServerServiceDefinition serviceStack = ServerInterceptors.intercept(
          new EchoTestService(),
          HttpHeaderInterceptors.serverInterceptor(),
          new GrpcLoggingInterceptor(),
          new MyAuthenticationInterceptor() // your interceptor stack.
  );

The client interceptor is automatically attached to your stubs by code generation after grpc-jersey 0.3.0.

To read & manipulate HTTP headers, below is an example right from the EchoTestService in this project:

    @Override
    public void testMethod3(TestRequest request, StreamObserver<TestResponse> responseObserver) {
        for (Map.Entry<String, String> header : HttpHeaderContext.requestHeaders().entries()) {
            if (header.getKey().startsWith("grpc-jersey")) {
                HttpHeaderContext.setResponseHeader(header.getKey(), header.getValue());
            }
        }

        responseObserver.onNext(TestResponse.newBuilder().setRequest(request).build());
        responseObserver.onCompleted();
    }

HttpHeaderContext is your interface into the HTTP headers. You can see request headers with HttpHeaderContext.requestHeaders() and set response headers with HttpHeaderContext.setResponseHeader(headerName, headerValue) or HttpHeaderContext.addResponseHeader(headerName, headerValue), the former setting a single value (or list of values), clearing existing ones, and the latter adding values. You can use HttpHeaderContext.clearResponseHeader(headerName) or HttpHeaderContext.clearResponseHeaders() to remove header state. Note: this API is considered beta and may change in the future.

While HttpHeaderContext is gRPC Context-aware and request headers can be safely accessed from background threads executed with an attached context, manipulating response headers should only be done from a single thread as no effort is put into synchronizing the state.

Streaming RPCs will apply any headers set before the first message is sent or before the stream is closed if no messages are sent.

Headers are optionally applied by the GrpcJerseyErrorHandler on error (for unary RPCs). The default (provided) implementation will honor headers set by the RPC handler on all error responses.

Additionally, as per the caveats with streaming RPCs in general, any additional headers added to an in-progress stream will be ignored, as headers can only be sent once in HTTP/1.x's common implementations, only headers present before the first message (or close/error) will be applied.

HTTP request headers are read into the main gRPC Metadata when using the "proxy" mode by default, however this is considered deprecated behavior. Utilizing the new HttpHeaderContext is the supported method.

grpc-jersey will translate errors raised inside your RPC handler. However, there is some nuance with regards to using the "proxy" mode or the "direct invocation" mode. If you use the direct invocation on service implementation, uncaught exceptions will use the default Jersey error handler. If you use the proxy mode, uncaught exceptions will be translated as INTERNAL errors.

Errors are translated into a google.rpc.Status message, which has the format when translated to JSON:

{
   "code":15,
   "message":"HTTP 500 (gRPC: DATA_LOSS): Fail-fast: Grue found in write-path.",
   "details":[

   ]
}

This payload will be returned in lieu of the actual return type.

The translation of gRPC error code to HTTP error code is done on a best-effort basis:

The HTTP status code will be applied to the outgoing response as the status code, but will also be a part of the message as seen above. The rest of the message comes from the gRPC StatusException/StatusRuntimeException description set by withDescription(String). Augmenting the description will append newlines which will be escaped in the final output.

The details section will always be empty. Protobuf JsonFormat does not support serializing the google.protobuf.Any type. Any details provided will be stripped by the error handling.

If google.rpc.RetryInfo is provided in the details section, this will be translated into a Retry-After header, e.x:

Metadata metadata = new Metadata();
metadata.put(GrpcErrorUtil.RETRY_INFO_KEY,
        RetryInfo.newBuilder().setRetryDelay(Durations.fromSeconds(30)).build());
responseObserver.onError(
        Status.RESOURCE_EXHAUSTED
        .asRuntimeException(metadata));

$ curl -v http://localhost:8080/explode
< HTTP/1.1 503 Service Unavailable
< Retry-After: 30
< Content-Type: application/json;charset=UTF-8
<
{
  "code": 8,
  "message": "HTTP 503 (gRPC: RESOURCE_EXHAUSTED)"
}

Streaming RPCs have to be handled a little differently. Headers are sent immediately before responses are produced, and streaming RPCs could fail at any point during streaming. In gRPC, this is handled with a trailer (like a header, but after the response is produced), but trailers are rarely if ever supported in HTTP/1.1 and often not even in HTTP2. Therefore, grpc-jersey signals failure by emitting a google.protobuf.Status should your handler return an error at any point during streaming. For instance:

> GET /stream/grpc_data_loss?int3=3 HTTP/1.1
>
< HTTP/1.1 200 OK
< Content-Type: application/json;charset=utf-8
< Transfer-Encoding: chunked
<
{"request":{"s":"grpc_data_loss","uint3":0,"uint6":"0","int3":3,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}
{"request":{"s":"grpc_data_loss","uint3":0,"uint6":"0","int3":3,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}
{"request":{"s":"grpc_data_loss","uint3":0,"uint6":"0","int3":3,"int6":"0","bytearray":"","boolean":false,"f":0.0,"d":0.0,"enu":"FIRST","rep":[],"repStr":[]}}
{"code":15,"message":"HTTP 500 (gRPC: DATA_LOSS): Fail-fast: Grue found in write-path.\ntest","details":[]}

This behavior can be overridden if desired.

If your project needs to handle errors differently (e.g. you have a standard error payload already, want to change error codes mappings, change streaming errors, etc.) you can override error handling at a JVM-global level.

During initialization of your project (RPC server setup), you can provide an implementation of a GrpcJerseyErrorHandler, see the Default implementation.

ErrorHandler.setErrorHandler(new MyGrpcJerseyErrorHandler());

JSON serialization/deserialization is done with protobuf's JsonFormat. By default, grpc-jersey emits all fields, even if they're set to their default value/empty. This assists with frontends that wish to traverse through structures.

Unary RPCs are emitted with formatting by default, but streaming RPCs are emitted on a single line.

Like error handlers, JSON formatters can be swapped out on a JVM-global basis.

JsonHandler.setParser(JsonFormat.parser());
JsonHandler.setUnaryPrinter(JsonFormat.printer());
JsonHandler.setStreamPrinter(JsonFormat.printer());

This can be used to disable emitting default fields, change formatting, or set parser/printers with ExtensionRegistry instances.

0.3.1

• Fix thread-safety issue with Context default values. Thanks @smartwjw. #27

0.3.0

• First-class HTTP header support. HTTP request headers are read into and attached to the gRPC Context. Likewise, response headers can be controlled from within your RPC handler. See Working with HTTP headers. #23
• Breaking change: API of GrpcJerseyErrorHandler has changed. If you haven't implemented a custom error handler, this doesn't affect you. If you have, please migrate your handler to the new API.

0.2.0

• Server-to-client RPC streaming support. #14
• ALREADY_EXISTS gRPC error code now maps to 409 Conflict.
• Error handling is now pluggable. See Overriding Error Handling.
• JSON printer/parser is now pluggable. See Overrding JSON formatting.
• 'Proxy' mode is now default code generation mode. #19
• Updated to protobuf 3.5, gRPC 1.8.
• More documentation! Added dual-stack server example.
• More integration testing around error handling.

0.1.4

• Changed to 'com.xorlev' artifact group, released on Sonatype/Central.
• Query parameters now support repeated types. @gfecher (#15)
• Windows artifact is now generated. @gfecher (#16)

0.1.3

• YAML support for defining resources and driving code generation. @sypticus (#10, #12

ALPHA

This project is in use and under active development.

Short-term roadmap:

• Documentation
• Support recursive path expansion for path parameters
• Support recursive path expansion for query parameters
• Support recursive path expansion for body parameters
• additional_bindings support
• Support for wildcard * and ** anonymous/named path expansion
• Support for endpoint definitions in a .yml file.
• response_body support
• Performance tests
• Generic/pluggable error handling
• Supporting streaming RPCs
  • Server streaming
  • Client streaming
  • BiDi streaming (true bidi streaming is impossible without websockets)
• Direct control of HTTP headers
• Out of the box CORS support
• Better deadline handling

Long-term roadmap:

• Potentially replace Jersey resources with servlet filter. This would make streaming easier.

./gradlew clean build

Please use --no-ff when merging feature branches.