pw_rpc can generate services which encode/decode RPC requests and responses as nanopb message structs.


To enable nanopb code generation, the build argument dir_pw_third_party_nanopb must be set to point to a local nanopb installation. Nanopb 0.4 is recommended, but Nanopb 0.3 is also supported.

Define a pw_proto_library containing the .proto file defining your service (and optionally other related protos), then depend on the nanopb_rpc version of that library in the code implementing the service.

# chat/


pw_proto_library("chat_protos") {
  sources = [ "chat_protos/chat_service.proto" ]

# Library that implements the Chat service.
pw_source_set("chat_service") {
  sources = [
  public_deps = [ ":chat_protos.nanopb_rpc" ]

A C++ header file is generated for each input .proto file, with the .proto extension replaced by .rpc.pb.h. For example, given the input file chat_protos/chat_service.proto, the generated header file will be placed at the include path "chat_protos/chat_service.rpc.pb.h".

Generated code API

All examples in this document use the following RPC service definition.

// chat/chat_protos/chat_service.proto

syntax = "proto3";

service Chat {
  // Returns information about a chatroom.
  rpc GetRoomInformation(RoomInfoRequest) returns (RoomInfoResponse) {}

  // Lists all of the users in a chatroom. The response is streamed as there
  // may be a large amount of users.
  rpc ListUsersInRoom(ListUsersRequest) returns (stream ListUsersResponse) {}

  // Uploads a file, in chunks, to a chatroom.
  rpc UploadFile(stream UploadFileRequest) returns (UploadFileResponse) {}

  // Sends messages to a chatroom while receiving messages from other users.
  rpc Chat(stream ChatMessage) returns (stream ChatMessage) {}


A C++ class is generated for each service in the .proto file. The class is located within a special pw_rpc::nanopb sub-namespace of the file’s package.

The generated class is a base class which must be derived to implement the service’s methods. The base class is templated on the derived class.

#include "chat_protos/chat_service.rpc.pb.h"

class ChatService final : public pw_rpc::nanopb::Chat::Service<ChatService> {
  // Implementations of the service's RPC methods; see below.

Unary RPC

A unary RPC is implemented as a function which takes in the RPC’s request struct and populates a response struct to send back, with a status indicating whether the request succeeded.

pw::Status GetRoomInformation(pw::rpc::
                              const RoomInfoRequest& request,
                              RoomInfoResponse& response);

Server streaming RPC

A server streaming RPC receives the client’s request message alongside a ServerWriter, used to stream back responses.

void ListUsersInRoom(pw::rpc::
                     const ListUsersRequest& request,
                     pw::rpc::ServerWriter<ListUsersResponse>& writer);

The ServerWriter object is movable, and remains active until it is manually closed or goes out of scope. The writer has a simple API to return responses:

Status ServerWriter::Write(const T &response)

Writes a single response message to the stream. The returned status indicates whether the write was successful.

void ServerWriter::Finish(Status status = OkStatus())

Closes the stream and sends back the RPC’s overall status to the client.

Once a ServerWriter has been closed, all future Write calls will fail.


Make sure to use std::move when passing the ServerWriter around to avoid accidentally closing it and ending the RPC.

Client streaming RPC


pw_rpc does not yet support client streaming RPCs.

Bidirectional streaming RPC


pw_rpc does not yet support bidirectional streaming RPCs.


A corresponding client class is generated for every service defined in the proto file. To allow multiple types of clients to exist, it is placed under the pw_rpc::nanopb namespace. The Client class is nested under pw_rpc::nanopb::ServiceName. For example, the Chat service would create pw_rpc::nanopb::Chat::Client.

Service clients are instantiated with a reference to the RPC client through which they will send requests, and the channel ID they will use.

// Nested under pw_rpc::nanopb::ServiceName.
class Client {
  Client(::pw::rpc::Client& client, uint32_t channel_id);

  GetRoomInformationCall GetRoomInformation(
      const RoomInfoRequest& request,
      ::pw::Function<void(Status, const RoomInfoResponse&)> on_response,
      ::pw::Function<void(Status)> on_rpc_error = nullptr);

  // ...and more (see below).

RPCs can also be invoked individually as free functions:

GetRoomInformationCall call = pw_rpc::nanopb::Chat::GetRoomInformation(
    client, channel_id, request, on_response, on_rpc_error);

The client class has member functions for each method defined within the service’s protobuf descriptor. The arguments to these methods vary depending on the type of RPC. Each method returns a NanopbClientCall object which stores the context of the ongoing RPC call. For more information on ClientCall objects, refer to the core RPC docs. The type of the returned object is complex, so it is aliased using the method name.

Callback invocation

RPC callbacks are invoked synchronously from Client::ProcessPacket.

Method APIs

The arguments provided when invoking a method depend on its type.

Unary RPC

A unary RPC call takes the request struct and a callback to invoke when a response is received. The callback receives the RPC’s status and response struct.

An optional second callback can be provided to handle internal errors.

GetRoomInformationCall GetRoomInformation(
    const RoomInfoRequest& request,
    ::pw::Function<void(const RoomInfoResponse&, Status)> on_response,
    ::pw::Function<void(Status)> on_rpc_error = nullptr);
Server streaming RPC

A server streaming RPC call takes the initial request struct and two callbacks. The first is invoked on every stream response received, and the second is invoked once the stream is complete with its overall status.

An optional third callback can be provided to handle internal errors.

ListUsersInRoomCall ListUsersInRoom(
    const ListUsersRequest& request,
    ::pw::Function<void(const ListUsersResponse&)> on_response,
    ::pw::Function<void(Status)> on_stream_end,
    ::pw::Function<void(Status)> on_rpc_error = nullptr);

Example usage

The following example demonstrates how to call an RPC method using a nanopb service client and receive the response.

#include "chat_protos/chat_service.rpc.pb.h"

namespace {

  using ChatClient = pw_rpc::nanopb::Chat::Client;

  MyChannelOutput output;
  pw::rpc::Channel channels[] = {pw::rpc::Channel::Create<1>(&output)};
  pw::rpc::Client client(channels);

  // Callback function for GetRoomInformation.
  void LogRoomInformation(const RoomInfoResponse& response, Status status);

}  // namespace

void InvokeSomeRpcs() {
  // Instantiate a service client to call Chat service methods on channel 1.
  ChatClient chat_client(client, 1);

  // The RPC will remain active as long as `call` is alive.
  auto call = chat_client.GetRoomInformation(
      {.room = "pigweed"}, LogRoomInformation);
  if (! {
    // The invocation may fail. This could occur due to an invalid channel ID,
    // for example. The failure status is forwarded to the to call's
    // on_rpc_error callback.

  // For simplicity, block until the call completes. An actual implementation
  // would likely std::move the call somewhere to keep it active while doing
  // other work.
  while ( {

  // Do other stuff now that we have the room information.


To enable pw_rpc.nanopb.* for Zephyr add CONFIG_PIGWEED_RPC_NANOPB=y to the project’s configuration. This will enable the Kconfig menu for the following:

  • pw_rpc.nanopb.method which can be enabled via CONFIG_PIGWEED_RPC_NANOPB_METHOD=y.

  • pw_rpc.nanopb.method_union which can be enabled via CONFIG_PIGWEED_RPC_NANOPB_METHOD_UNION=y.

  • pw_rpc.nanopb.client which can be enabled via CONFIG_PIGWEED_RPC_NANOPB_CLIENT=y.

  • pw_rpc.nanopb.common which can be enabled via CONFIG_PIGWEED_RPC_NANOPB_COMMON=y.

  • pw_rpc.nanopb.echo_service which can be enabled via CONFIG_PIGWEED_RPC_NANOPB_ECHO_SERVICE=y.