コンテンツにスキップ
|
Plugin Hub
report-issue問題を報告する
header icon

Getting started with gRPC-Web

By Kong Inc.

このページは、まだ日本語ではご利用いただけません。翻訳中です。

This plugin should be enabled on a Kong route that serves the HTTP(S) protocol but proxies to a service with the GRPC(S) protocol.

Sample configuration via declarative (YAML):

_format_version: "3.0"
services:
- protocol: grpc
  host: localhost
  port: 9000
  routes:
  - protocols:
    - name: http
    paths:
    - /
    plugins:
    - name: grpc-web

Same thing via the Admin API:

  1. Add the gRPC service: 
      curl -X POST localhost:8001/services \
 --data name=grpc \
 --data protocol=grpc \
 --data host=localhost \
 --data port=9000
  2. Add an http route: 
      curl -X POST localhost:8001/services/grpc/routes \
 --data protocols=http \
 --data name=web-service \
 --data paths[]=/
  3. Add the plugin to the route: 
      curl -X POST localhost:8001/routes/web-service/plugins \
 --data name=grpc-web

In these examples, we don’t set any plugin configurations. This minimal setup works for the default varieties of the gRPC-Web protocol, which use ProtocolBuffer messages either directly in binary or with base64-encoding. The related Content-Type headers are application/grpc-web or application/grpc-web+proto for binary, and application/grpc-web-text or application/grpc-web-text+proto for text.

If you want to use JSON encoding, you have to provide the gRPC specification in a .proto file, which needs to be installed in the Kong node running the plugin. A path starting with a / is considered absolute; otherwise, it interprets relative to the Kong node’s prefix (/usr/local/kong/ by default).

Example Protobuf definition (hello.proto):

syntax = "proto2";

package hello;

service HelloService {
  rpc SayHello(HelloRequest) returns (HelloResponse);
  rpc LotsOfReplies(HelloRequest) returns (stream HelloResponse);
  rpc LotsOfGreetings(stream HelloRequest) returns (HelloResponse);
  rpc BidiHello(stream HelloRequest) returns (stream HelloResponse);
}

message HelloRequest {
  optional string greeting = 1;
}

message HelloResponse {
  required string reply = 1;
}

The declarative configuration becomes:

_format_version: "3.0"
services:
- protocol: grpc
  host: localhost
  port: 9000
  routes:
  - protocols:
    - name: http
    paths:
    - /
    plugins:
    - name: grpc-web
      proto: path/to/hello.proto

or via the Admin API:

curl -X POST localhost:8001/routes/web-service/plugins \
  --data name=grpc-web \
  --data config.proto=path/to/hello.proto

Upload the Protobuf definition to your Kong Node:

docker cp hello.proto localhost:/usr/local/kong/

Test your setup:

curl -X POST localhost:8000/hello.HelloService/SayHello \
--header 'x-grpc: true' \
--header 'Content-Type: application/json' \
--data '{"greeting":"kong2.1"}'

With this setup, there’s support for gRPC-Web/JSON clients using Content-Type headers like application/grpc-web+json or application/grpc-web-text+json.

Note: When using JSON encoding, the gRPC-Web protocol specifies that both request and response data consist of a series of frames, in a similar way to the full gRPC protocol. The gRPC-Web library performs this framing as expected.

As an extension, this plugin also allows naked JSON requests with the POST method and Content-Type: application/json header. These requests are encoded to ProtocolBuffer, framed, and forwarded to the gRPC service. Likewise, the responses are transformed on the way back, allowing any HTTP client to use a gRPC service without special libraries. This feature is limited to unary (non-streaming) requests. Streaming responses are encoded into multiple JSON objects; it’s up to the client to split into separate records if it has to support multiple response messages.