gRPC效能比較

gPRC壓力測試

gRPC 有支援的程式語言很多，有 Python、golang、js、java 等等，所以以下會比較go和python兩個language的implementation performance

ghz-壓力測試工具

Simple gRPC benchmarking and load testing tool inspired by hey and grpcurl.
github
documentation

Installation

• window
  去這裡下載解壓縮就可以了
• linux

  wget https://github.com/bojand/ghz/releases/download/v0.52.0/ghz_0.52.0_Linux_x86_64.tar.gz
  sudo tar -xvf ghz_0.52.0_Linux_x86_64.tar.gz

  Additionally consider adding the location directory in the PATH variable if you would like the ghz command to be available everywhere.

Usage

usage: ghz [<flags>] [<host>]

Flags:
  -h, --help                   Show context-sensitive help (also try --help-long and --help-man).
      --config=                Path to the JSON or TOML config file that specifies all the test run settings.
      --proto=                 The Protocol Buffer .proto file.
      --protoset=              The compiled protoset file. Alternative to proto. -proto takes precedence.
      --call=                  A fully-qualified method name in 'package.Service/method' or 'package.Service.Method' format.
  -i, --import-paths=          Comma separated list of proto import paths. The current working directory and the directory of the protocol buffer file are automatically added to the import list.
      --cacert=                File containing trusted root certificates for verifying the server.
      --cert=                  File containing client certificate (public key), to present to the server. Must also provide -key option.
      --key=                   File containing client private key, to present to the server. Must also provide -cert option.
      --cname=                 Server name override when validating TLS certificate - useful for self signed certs.
      --skipTLS                Skip TLS client verification of the server's certificate chain and host name.
      --insecure               Use plaintext and insecure connection.
      --authority=             Value to be used as the :authority pseudo-header. Only works if -insecure is used.
  -c, --concurrency=50         Number of requests to run concurrently. Total number of requests cannot be smaller than the concurrency level. Default is 50.
  -n, --total=200              Number of requests to run. Default is 200.
  -q, --qps=0                  Rate limit, in queries per second (QPS). Default is no rate limit.
  -t, --timeout=20s            Timeout for each request. Default is 20s, use 0 for infinite.
  -z, --duration=0             Duration of application to send requests. When duration is reached, application stops and exits. If duration is specified, n is ignored. Examples: -z 10s -z 3m.
  -x, --max-duration=0         Maximum duration of application to send requests with n setting respected. If duration is reached before n requests are completed, application stops and exits. Examples: -x 10s -x 3m.
      --duration-stop="close"  Specifies how duration stop is reported. Options are close, wait or ignore.
  -d, --data=                  The call data as stringified JSON. If the value is '@' then the request contents are read from stdin.
  -D, --data-file=             File path for call data JSON file. Examples: /home/user/file.json or ./file.json.
  -b, --binary                 The call data comes as serialized binary message or multiple count-prefixed messages read from stdin.
  -B, --binary-file=           File path for the call data as serialized binary message or multiple count-prefixed messages.
  -m, --metadata=              Request metadata as stringified JSON.
  -M, --metadata-file=         File path for call metadata JSON file. Examples: /home/user/metadata.json or ./metadata.json.
      --stream-interval=0      Interval for stream requests between message sends.
      --reflect-metadata=      Reflect metadata as stringified JSON used only for reflection request.
  -o, --output=                Output path. If none provided stdout is used.
  -O, --format=                Output format. One of: summary, csv, json, pretty, html, influx-summary, influx-details. Default is summary.
      --connections=1          Number of connections to use. Concurrency is distributed evenly among all the connections. Default is 1.
      --connect-timeout=10s    Connection timeout for the initial connection dial. Default is 10s.
      --keepalive=0            Keepalive time duration. Only used if present and above 0.
      --name=                  User specified name for the test.
      --tags=                  JSON representation of user-defined string tags.
      --cpus=12                Number of cpu cores to use.
      --debug=                 The path to debug log file.
  -e, --enable-compression     Enable Gzip compression on requests.
  -v, --version                Show application version.

Args:
  [<host>]  Host and port to test.

Example proto :

syntax = "proto3";  

package pb;

service HelloService {
  rpc SayHello (HelloRequest) returns (HelloResponse);
}

message HelloRequest {
  string greeting = 1;
}

message HelloResponse {
  string reply = 1;
}

We can test the call:

ghz --insecure \
  --proto ./hello.proto \
  --call  pb.HelloService.SayHello \
  -d '{"greeting":"hello world"}' \
  127.0.0.1:9999

--insecure : Use plaintext and insecure connection
proto: proto檔案路徑
call : proto 定義的RPC服務 <package.Service.Method>
-d : 傳入的參數
127.0.0.1:9999 : host gRPC service

在window 下實測後無法直在terminal用-d帶入參數，必須透過-D另外帶上json檔案路徑的方式不然會報錯

json example:

{"greeting":"hello world"}

command :

ghz --insecure \
  --proto ./hello.proto \
  --call  pb.HelloService.SayHello \
  -D task_data.json \
  127.0.0.1:9999

Advanced usage

• 可以將所有的設定寫成json格式，方便管理

json example:

{
    "total": 5000,
    "concurrency": 50,
    "proto": "./hello.proto",
    "call": "pb.HelloService.SayHello",
    "data": {
      "greeting": "hello world"
    },
    "host": "127.0.0.1:9999",
    "insecure": true
  }

command :

ghz.exe --config .\config.json

• 支援多種輸出格式，如csv, json, html, InfluxDB

export html report command :

ghz.exe --insecure \
--proto ./hello.proto \
--call pb.HelloService.SayHello \
-D task_data.json \
-c 1000 -n 10000 -O html \
127.0.0.1:9998 >> report.html

Experiment

python gRPC server:127.0.0.1:9999

from concurrent import futures
import time

import grpc
import hello_pb2
import hello_pb2_grpc


# 創建一個 HelloServiceServicer，要繼承自 hello_pb2_grpc.HelloServiceServicer
class HelloServiceServicer(hello_pb2_grpc.HelloServiceServicer):

# 由於我們 service 定義了 SayHello 這個 rpc，所以要實作 SayHello 這個 method
    def SayHello(self, request, context):
# response 是個 HelloResponse 形態的 message
        response = hello_pb2.HelloResponse()
        response.reply = f'Hello, {request.greeting}'
        return response


def serve():
# 創建一個 gRPC server
    server = grpc.server(futures.ThreadPoolExecutor())

# 利用 add_HelloServiceServicer_to_server 這個 method 把上面定義的 HelloServiceServicer 加到 server 當中
    hello_pb2_grpc.add_HelloServiceServicer_to_server(HelloServiceServicer(), server)

# 讓 server 跑在 port 9999 中
    server.add_insecure_port('[::]:9999')
    server.start()
    try:
        while True:
            time.sleep(86400)
    except KeyboardInterrupt:
        server.stop(0)


if __name__ == '__main__':
    serve()

go gRPC server:127.0.0.1:9998

package main

import (
	"context"
	"log"
	"net"

	pb "../pb"
	"google.golang.org/grpc"
)

type service struct {
	pb.UnimplementedHelloServiceServer
}

func (s *service) SayHello(ctx context.Context, in *pb.HelloRequest) (*pb.HelloResponse, error) {
	// log.Printf("Received: %v", in.GetGreeting())
	return &pb.HelloResponse{Reply: "Hello, " + in.GetGreeting()}, nil
}

func main() {
	addr := "127.0.0.1:9998"
	lis, err := net.Listen("tcp", addr)
	if err != nil {
		log.Fatalf("failed to listen: %v", err)
	}

	// log.Println("Server listening on", addr)
	gRPCServer := grpc.NewServer()
	pb.RegisterHelloServiceServer(gRPCServer, &service{})
	if err := gRPCServer.Serve(lis); err != nil {
		log.Fatalf("failed to serve: %v", err)
	}
}

上面的code都放在這邊

透過ghz tool 對python server進行1000000個request，同時1000個連線，且每次都傳1kB的資料

ghz.exe --insecure \
--import-paths ./ \
--proto ./hello.proto \
--call pb.HelloService.SayHello \
--keepalive 3s \
-D task_data.json \
-c 1000 -n 1000000 \
127.0.0.1:9999

透過ghz tool 對go server進行1000000個request，同時1000個連線，且每次都傳1kB的資料

ghz.exe --insecure \
--import-paths ./ \
--proto ./hello.proto \
--call pb.HelloService.SayHello \
--keepalive 3s \
-D task_data.json \
-c 1000 -n 1000000 \
127.0.0.1:9998

可以發現 Reqursts/sec go為63200.71 而python 則比較少為5963.62，所以go的Requrests/sec大約為python的10倍