gRPC背压流控、压缩及JSON通信【知识笔记】

一、压缩
    1.Server端所有方法压缩
    2.Server单独方法压缩
    3.Client请求内容压缩
二、使用JSON通信
    1.方法描述使用JSON编译
    2.JSON编译具体过程
三、手动流量控制
    1.Consuming Side
    2.Producing Side
四、系列文章

server = ServerBuilder.forPort(port)
.intercept(new ServerInterceptor() {
  @Override
  public <ReqT, RespT> Listener<ReqT> interceptCall(ServerCall<ReqT, RespT> call, Metadata headers,
      ServerCallHandler<ReqT, RespT> next) {
    // @1 在拦截器中设置压缩算法
    call.setCompression("gzip");
    return next.startCall(call, headers);
  }
})
.addService(new GreeterImpl())
.build()
.start();

int port = 50051;
server = ServerBuilder.forPort(port)
    .addService(new GreeterImpl())
    .build()
    .start();
static classGreeterImplextendsGreeterGrpc.GreeterImplBase{

@Override
public void sayHello(HelloRequest req, StreamObserver<HelloReply> plainResponseObserver) {
  ServerCallStreamObserver<HelloReply> responseObserver =
      (ServerCallStreamObserver<HelloReply>) plainResponseObserver;
  // @1 对单个方法传输内容进行压缩
  responseObserver.setCompression("gzip");
  HelloReply reply = HelloReply.newBuilder().setMessage("Hello " + req.getName()).build();
  responseObserver.onNext(reply);
  responseObserver.onCompleted();
}
}

publicvoidgreet(String name) {
HelloRequest request = HelloRequest.newBuilder().setName(name).build();
HelloReply response;
try {
  // @1 对请求内容设置压缩类型
  response = blockingStub.withCompression("gzip").sayHello(request);
} catch (StatusRuntimeException e) {
  logger.log(Level.WARNING, "RPC failed: {0}", e.getStatus());
  return;
}

static final MethodDescriptor<HelloRequest, HelloReply> METHOD_SAY_HELLO =
GreeterGrpc.getSayHelloMethod()
    .toBuilder(
 // @1 请求参数使用JSON编译   JsonMarshaller.jsonMarshaller(HelloRequest.getDefaultInstance()),
// @2 返回参数使用JSON编译
 JsonMarshaller.jsonMarshaller(HelloReply.getDefaultInstance()))
    .build();

public static <T extends Message> Marshaller<T> jsonMarshaller(final T defaultInstance) {
    final Parser parser = JsonFormat.parser();
    final Printer printer = JsonFormat.printer();
    return jsonMarshaller(defaultInstance, parser, printer);
}
public static <T extends Message> Marshaller<T> jsonMarshaller(
      final T defaultInstance, final Parser parser, final Printer printer) {
final Charset charset = Charset.forName("UTF-8");
return new Marshaller<T>() {
  @Override
  public InputStream stream(T value) {
    try {
      // @1 通过printer.print将出入参数转换为JSON格式
      return new ByteArrayInputStream(printer.print(value).getBytes(charset));
    } catch (InvalidProtocolBufferException e) {
      // ...
    }
  }
// ....
}

public HelloReply sayHello(HelloRequest request) {
 // @1 使用JSON格式的方法描述METHOD_SAY_HELLO
  return blockingUnaryCall(
      getChannel(), METHOD_SAY_HELLO, getCallOptions(), request);
}

public ServerServiceDefinition bindService() {
  return ServerServiceDefinition
      .builder(GreeterGrpc.getServiceDescriptor().getName())
       // @1 使用JSON格式的方法描述METHOD_SAY_HELLO
      .addMethod(HelloJsonClient.HelloJsonStub.METHOD_SAY_HELLO,
          asyncUnaryCall(
              new UnaryMethod<HelloRequest, HelloReply>() {
                @Override
                publicvoidinvoke(
                    HelloRequest request, StreamObserver<HelloReply> responseObserver) {
                  sayHello(request, responseObserver);
                }
              }))
      .build();
}

At the bottom is the HTTP/2's byte-based flow control. HTTP/2 works on streams of bytes and is completely unaware of gRPC messages or reactive streams. By default, the stream consumer allocates a budget of 65536 bytes.
The stream producer can send up to this many bytes before backpressure engages. As the consumer reads bytes, WINDOW_UPDATE messages are sent to the producer to increase its send budget.

In the middle is the gRPC-Java message-based flow control. gRPC's flow control adapts the stream-based flow control of HTTP/2 to a message-based flow control model.
Importantly, gRPC's flow control is aware of how it interacts with HTTP/2 and the network.

On producing side, an on-ready handler reads a message, serializes it into bytes using protobuf, and then queues it up for transmission over the HTTP/2 byte stream.
If there is insuficient room in the HTTP/2 flow control window to transmit, backpressure engages an no more messages are requested from the producer until space becomes available.

On the consuming side, each time a consumer calls request(x), gRPC attempts to read and deserialize x messages from the HTTP/2 stream.
Since the size of a protobuf encoded message is variable, there is not a one-to-one correlation between pulling messages from gRPC and pulling bytes over HTTP/2.

publicstaticvoidmain(String[] args) throws InterruptedException, IOException {
// @1 Server端服务实现
StreamingGreeterGrpc.StreamingGreeterImplBase svc = new StreamingGreeterGrpc.StreamingGreeterImplBase() {
  @Override
  public StreamObserver<HelloRequest> sayHelloStreaming(final StreamObserver<HelloReply> responseObserver) {
    final ServerCallStreamObserver<HelloReply> serverCallStreamObserver =
        (ServerCallStreamObserver<HelloReply>) responseObserver;
    // @2 禁止自动流控模式，开启手动流控
    serverCallStreamObserver.disableAutoInboundFlowControl();
    // @3 背压模式流控，当消费端有足够空间时将会回调OnReadyHandler
    // 默认空间大小为65536字节
     classOnReadyHandlerimplementsRunnable{
      private boolean wasReady = false;

      @Override
      publicvoidrun() {
        if (serverCallStreamObserver.isReady() && !wasReady) {
          wasReady = true;
          logger.info("READY");
          // @4 向HTTP/2流请求读取并解压(x)条消息
          // 即发信号通知发送端发送继续发消息
          serverCallStreamObserver.request(1);
        }
      }
    }
    final OnReadyHandler onReadyHandler = new OnReadyHandler();
    serverCallStreamObserver.setOnReadyHandler(onReadyHandler);
    // @5 处理具体进来的请求
    return new StreamObserver<HelloRequest>() {
      @Override
      publicvoidonNext(HelloRequest request) {
        try {
          String name = request.getName();
          logger.info("--> " + name);
          Thread.sleep(100);
          String message = "Hello " + name;
          logger.info("<-- " + message);
          HelloReply reply = HelloReply.newBuilder().setMessage(message).build();
          // @6 向Client发送请求
          responseObserver.onNext(reply);
          if (serverCallStreamObserver.isReady()) {
            // @7 向HTTP/2流请求读取并解压(x)条消息
            serverCallStreamObserver.request(1);
          } else {
            onReadyHandler.wasReady = false;
          }
        } catch (Throwable throwable) {
          //
        }
      }
      @Override
      publicvoidonError(Throwable t) {
        t.printStackTrace();
        responseObserver.onCompleted();
      }
      @Override
      publicvoidonCompleted() {
        logger.info("COMPLETED");
        responseObserver.onCompleted();
      }
    };
  }
};

final Server server = ServerBuilder
    .forPort(50051)
    .addService(svc)
    .build()
    .start();

publicstaticvoidmain(String[] args) throws InterruptedException {
final CountDownLatch done = new CountDownLatch(1);
ManagedChannel channel = ManagedChannelBuilder
    .forAddress("localhost", 50051)
    .usePlaintext()
    .build();
StreamingGreeterGrpc.StreamingGreeterStub stub = StreamingGreeterGrpc.newStub(channel);

ClientResponseObserver<HelloRequest, HelloReply> clientResponseObserver =
    new ClientResponseObserver<HelloRequest, HelloReply>() {
      ClientCallStreamObserver<HelloRequest> requestStream;
      @Override
      publicvoidbeforeStart(final ClientCallStreamObserver<HelloRequest> requestStream) {
        this.requestStream = requestStream;
        // @1设置手动流量控制
        requestStream.disableAutoInboundFlowControl();
        // @2 当Consumer端有足够空间时自动回调
        // 序列化protobuf先发送到缓存区（还未到Server端）
        // Server端需要调用request()向Client拉取消息
        requestStream.setOnReadyHandler(new Runnable() {
        Iterator<String> iterator = names().iterator();
          @Override
          publicvoidrun() {
            while (requestStream.isReady()) {
              if (iterator.hasNext()) {
                  String name = iterator.next();
                  logger.info("--> " + name);
                  HelloRequest request = HelloRequest.newBuilder().setName(name).build();
                  // @3 将消息发送到缓存区
                  requestStream.onNext(request);
              } else {
                  // @4 标记Client发送完成
                  requestStream.onCompleted();
              }
            }
          }
        });
      }

      @Override
      publicvoidonNext(HelloReply value) {
        // @5 接受Server端返回信息
        logger.info("<-- " + value.getMessage());
        // @6 通知Client继续发送
        requestStream.request(1);
      }

      @Override
      publicvoidonError(Throwable t) {
        t.printStackTrace();
        done.countDown();
      }

      @Override
      publicvoidonCompleted() {
        logger.info("All Done");
        done.countDown();
      }
    };
stream processing.
stub.sayHelloStreaming(clientResponseObserver);

done.await();

channel.shutdown();
channel.awaitTermination(1, TimeUnit.SECONDS);
}