Client Lifecycle

Instances of the generated client may be long-lived and can be used concurrently. You can keep the client running until your system terminates, or close it earlier. To avoid leaking in the latter case, you should call .close() on the client.

When the connection breaks, the client will try reconnecting to the server automatically. On each reconnection attempt, If a connection the ServiceDiscovery will be used and a new host may be found.

When using client-side load balancing the reconnection loop will run indefinitely.

When using a direct client (not load balanced) when the connection breaks you can set up a maximum number of reconnection attempts. If that limit is reached, the client will shutdown. The default number of attempts to reconnect is infinite and configurable via GrpcClientSettings’s connectionAttempts.

The client offers a method closed() that returns a FutureCompletionStage that will complete once the client is explicitly closed after invoking close(). The returned FutureCompletionStage will complete with a failure when the maximum number of connectionAttempts (which causes a shutdown).

Shared Channels

By default, each instance of a generated client creates a separate HTTP connection to the server. If the server supports multiple services, you may want to allow multiple generated clients to share a single connection.

To do this, create a GrpcChannelGrpcChannel by passing GrpcClientSettingsGrpcClientSettings to the apply method. You can then use the GrpcChannel instance to create multiple generated clients; each client will use the provided channel to communicate with the server.

When using a shared channel, the client lifecycle changes slightly. Like the generated client, GrpcChannel offers close and closed methods; these can be used to explicitly close the connection to the server and detect when the connection has been closed or shutdown due to errors, respectively. When you are done communicating with the server, you should call close on the channel, rather than the individual clients. Calling close on a generated client that was created with a shared channel will throw a GrpcClientCloseExceptionGrpcClientCloseException.

Load balancing

When multiple endpoints are discovered for a gRPC client, currently one is selected and used for all outgoing calls.

This approach, while naïve, in fact works well in many cases: when there are multiple nodes available to handle requests, a server-side load balancer is better-positioned to make decisions than any single client, as it can take into account information from multiple clients, and sometimes even lifecycle information (e.g. not forward requests to nodes that are scheduled to shut down).

Client-side load balancing is desirable when you are using the default static or the grpc-dns discovery mechanism. You can set the load-balancing-policy client configuration option to round_robin to enable the round_robin client-side load balancing strategy provided by grpc-java.

Note that load balancing is marked as experimental in grpc-java.

Client-side load balancing for other discovery mechanisms is not yet supported.

Request Metadata

Default request metadata, for example for authentication, can be provided through the GrpcClientSettingsGrpcClientSettings passed to the client when it is created, it will be the base metadata used for each request.

In some cases you will want to provide specific metadata to a single request, this can be done through the “lifted” client API, each method of the service has an empty parameter list version (.sayHello()) on the client returning a SingleResponseRequestBuilderSingleResponseRequestBuilder or StreamResponseRequestBuilderStreamResponseRequestBuilder.

After adding the required metadata the request is done by calling invoke with the request parameters.

Notice: method addHeader return a new object, you should use it like String or use it in the chain structure.

sourcedef singleRequestReply(): Unit = {"Performing request")
  val reply = client.sayHello().addHeader("key", "value").invoke(HelloRequest("Alice"))
  println(s"got single reply: ${Await.result(reply, 5.seconds).message}")
sourceprivate static void singleRequestReply(GreeterServiceClient client) throws Exception {
  HelloRequest request = HelloRequest.newBuilder().setName("Alice").build();
  CompletionStage<HelloReply> reply = client.sayHello()
      .addHeader("key", "value")
  System.out.println("got single reply: " + reply.toCompletableFuture().get(5, TimeUnit.SECONDS));

Rich error model

Beyond status codes you can also use the Rich error model. Currently there is no particular support for consuming such error objects (such as the ones based on the common protobuf, but you can obtain them ‘manually’.

Extract the StatusRuntimeException and parse the Rich error model to access code, message and details. Then find the details you are looking for based on their typeUrl and unpack them:

sourceval richErrorResponse = client.sayHello(HelloRequest("Bob")).failed.futureValue

richErrorResponse match {
  case ex: StatusRuntimeException =>
    val status: = StatusProto.fromStatusAndTrailers(ex.getStatus, ex.getTrailers)

    def fromJavaProto(javaPbSource: = = javaPbSource.getTypeUrl, value = javaPbSource.getValue)

    status.getDetails(0).getTypeUrl should be("")

    import LocalizedMessage.messageCompanion
    val customErrorReply: LocalizedMessage = fromJavaProto(status.getDetails(0)).unpack

    status.getCode should be(3)
    status.getMessage should be("What is wrong?")
    customErrorReply.message should be("The password!")
    customErrorReply.locale should be("EN")

  case ex => fail(s"This should be a StatusRuntimeException but it is ${ex.getClass}")
sourceHelloRequest request = HelloRequest.newBuilder().setName("Alice").build();
CompletionStage<HelloReply> response = client.sayHello(request);
StatusRuntimeException statusEx = response.toCompletableFuture().handle((res, ex) -> {
    return (StatusRuntimeException) ex;
}).get(); status = StatusProto.fromStatusAndTrailers(statusEx.getStatus(), statusEx.getTrailers());

assertEquals("", status.getDetails(0).getTypeUrl()); details = fromJavaProto(status.getDetails(0)).unpack(;

assertEquals(Status.INVALID_ARGUMENT.getCode().value(), status.getCode());
assertEquals("What is wrong?", status.getMessage());
assertEquals("The password!", details.message());
assertEquals("EN", details.locale());

Please look here how to create errors with such details on the server side.

Found an error in this documentation? The source code for this page can be found here. Please feel free to edit and contribute a pull request.