一、流式RPC的三种具体形式
流式RPC不同于简单RPC只有“单发单收“一种形式,而是可以分为三种不同的形式——“应答流式RPC”,“请求流式RPC”,“双向流式RPC”。对于这三种不同的形式,python有不同的请求及接收方式,下面就让我们来具体了解一下。(对于下面操作有疑问的同学可以去看上一期的内容)
首先接口协议是有区别的,我们来看三种形式的接口定义:
应答流式RPC:
rpc ListFeatures(Rectangle) returns (stream Feature) {}
请求流式RPC:
rpc RecordRoute(stream Point) returns (RouteSummary) {}
双向流式RPC:
rpc RouteChat(stream RouteNote) returns (stream RouteNote) {}
可以看到,请求和响应参数中流式内容的前面会有一个stream标识,代表这是一个流式的内容。应答流式RPC只有返回是流式的,请求流式RPC只有请求是流式的,而双向流式RPC请求和返回都是流式的。
一个包含接口的完整proto协议文件(route_guide.proto)内容如下:
syntax = "proto3";option java_multiple_files = true;option java_package = "io.grpc.examples.routeguide";option java_outer_classname = "RouteGuideProto";option objc_class_prefix = "RTG";package routeguide;service RouteGuide {rpc ListFeatures(Rectangle) returns (stream Feature) {}rpc RecordRoute(stream Point) returns (RouteSummary) {}rpc RouteChat(stream RouteNote) returns (stream RouteNote) {}}message Point {int32 latitude = 1;int32 longitude = 2;}message Rectangle {Point lo = 1;Point hi = 2;}message Feature {string name = 1;Point location = 2;}message RouteNote {Point location = 1;string message = 2;}message RouteSummary {int32 point_count = 1;int32 feature_count = 2;int32 distance = 3;int32 elapsed_time = 4;}
根据协议文件生成route_guide_pb2.py、route_guide_pb2_grpc.py两个必要的模块文件,然后就可以根据他们来创建客户端了。
二、客户端实现
1、应答流式RPC
应答流式RPC返回的内容为流式,一次请求,n次返回。我们可以用for循环来接收返回的内容:
def guide_list_features(stub):rectangle = route_guide_pb2.Rectangle(lo=route_guide_pb2.Point(latitude=400000000, longitude=-750000000),hi=route_guide_pb2.Point(latitude=420000000, longitude=-730000000))print("Looking for features between 40, -75 and 42, -73")features = stub.ListFeatures(rectangle)for feature in features:print("Feature called %s at %s" % (feature.name, feature.location))
2、请求流式RPC
请求流式RPC请求的内容为流式,n次请求,一次返回。我们可以用迭代器来发送若干份请求数据:
def generate_route(feature_list):for _ in range(0, 10):random_feature = feature_list[random.randint(0, len(feature_list) - 1)]print("Visiting point %s" % random_feature.location)yield random_feature.locationdef guide_record_route(stub):feature_list = route_guide_resources.read_route_guide_database()route_iterator = generate_route(feature_list)route_summary = stub.RecordRoute(route_iterator)print("Finished trip with %s points " % route_summary.point_count)print("Passed %s features " % route_summary.feature_count)print("Travelled %s meters " % route_summary.distance)print("It took %s seconds " % route_summary.elapsed_time)
其中route_iterator为一个迭代器。
3、双向流式RPC
双向流式RPC请求的内容为流式,返回内容也为流式,n次请求,n次返回。我们可以用迭代器来发送若干份请求数据,通过for循环来接收返回结果:
def generate_messages():messages = [make_route_note("First message", 0, 0),make_route_note("Second message", 0, 1),make_route_note("Third message", 1, 0),make_route_note("Fourth message", 0, 0),make_route_note("Fifth message", 1, 0),]for msg in messages:print("Sending %s at %s" % (msg.message, msg.location))yield msgdef guide_route_chat(stub):responses = stub.RouteChat(generate_messages())for response in responses:print("Received message %s at %s" %(response.message, response.location))
三、实际应用
在录音笔项目中,需要对转写后的文本进行分段语义整理,由于文本内容可能较多,服务端需要采用流式的方式进行接收,并通过流式的方式将结果返给客户端,于是这里采用了双向流式RPC形式的接口。
接口协议如下(仅为演示需要,只展示部分内容):
syntax = "proto3";package sogou.parrot.inner.semantic.v1;import "google/protobuf/duration.proto";import "record.proto";option go_package = "git.speech.sogou/semantic/v1;semantic";service discourse_understand{rpc UnderstandFullText(stream UnderstandFullTextRequest) returns(stream UnderstandFullTextResponse);}message UnderstandFullTextRequest{repeated SubSentence sub_sentences = 1;repeated sogou.parrot.record.v1.NonSpeechSoundInfo sound_infos = 2;repeated sogou.parrot.record.v1.AIMark ai_marks = 3;}message UnderstandFullTextResponse{UnderstandFullTextResult result = 2;}
实现客户端的关键代码如下:
def gen_iterator(request):for r in [request]:yield rdef get_understand_full_textresponse(stub, ai_marks, sound_infos, sub_sentences):request = UnderstandFullTextRequest()request.sub_sentences.extend(sub_sentences)request.sound_infos.extend(sound_infos)request.ai_marks.extend(ai_marks)request_iter = gen_iterator(request)try:resps = stub.UnderstandFullText(request_iter)for resp in resps:resp_str = json.dumps(json.loads(MessageToJson(resp)),indent=4, ensure_ascii=False)print(resp_str)except Exception as e:print (e)def run():ai_marks, sound_infos, sub_sentences = extract_data()with grpc.insecure_channel(sys.argv[2]) as channel:stub = discourse_understandStub(channel)print("-------------- UnderstandFullText --------------")get_understand_full_textresponse(stub, ai_marks, sound_infos, sub_sentences)if __name__ == '__main__':run()
https://mp.weixin.qq.com/s/Y2sHs_Sq4lB3hBhKGSvaNg
程序员如何用gRPC谈一场恋爱
原创neil腾讯技术工程2019-06-12
语: 本文以幽默诙谐的方式,介绍gRPC的4种client-server服务模式的开发实践及应用场景
前言:为什么要写这篇文章?
The best way to learn is to teach. gRPC的examples里的例子是一个简易的router,琐碎的业务代码很多,看起来比较绕。于是就自己写一个例子,看看自己是否都将这些知识点掌握了。
谈恋爱的过程,其实跟client-server服务模式非常像。单独讲这4种模式有些无聊,所以就尝试用一种尽量有趣的方式去介绍,顺便也可以作为某些男生的一份简单的恋爱指南。
gRPC client-server服务模式
这里先将重要的结论写出来,方便以后查阅,具体介绍见下文。找不到准确的中文来翻译这几种模式,就保留了英文。
0x1: A simple RPC
最简单的一发一收的client-server模型。这就是我们用得最多的模式
0x2:A client-to-server streaming RPC
client先建立长连接,然后发送多个request给server,server最后统一回一个rsp
应用场景:
agent上报CPU,内存等数据到server
客户端心跳
客户端并发调用细小粒度的接口。比如有5个后台接口A B C D E,客户端在不同页面,可以调用不同的接口组合。比如在个人页,就调用ABC;在动态页面,就调用CDE,后台都只会有一个rsp。这种模式的好处就是让后台可以将接口的粒度细化,客户端调用灵活,减少重复代码,提高复用率
0x3: A server-to-client streaming RPC
client先发一个请求到server,然后server不停的回包
应用场景:
股票app。客户端向服务端发送一个股票代码,服务端就把该股票的实时数据源源不断的返回给客户端
app的在线push。client先发请求到server注册,然后server就可以发在线push了
0x4: A Bidirectional streaming RPC
建立一个长连接,然后client和server可以随意收发数据
应用场景:
聊天机器人
有状态的游戏服务器进行数据交换。比如LOL,王者荣耀等竞技游戏,client和server之间需要非常频繁地交换数据
总结:除了一发一收的模式,client-to-server与server-to-client这两种模式,其实用双向RPC都可以做到。但是双向RPC的编码难度更高(后面例子中可以看到),异常处理也需要更仔细,所以具体使用哪种模式需要根据具体需求来分析
接下来就让我用一个男女生之间的交往故事来说明这4种服务模式。内容纯属虚构,并故意写得比较搞笑。如果有不恰当的描述,请告诉我。
男女生交往之gRPC的4种模式
一些说明
Selina:女生名字,client
John:男生名字,server
等等,女生是client,男生是server。那么,女生给男生提要求,是不是就是client向server请求数据一样自然!!!!噢噢噢,这应该是我学编程以来,领悟到的最重要的道理吧!!!
基础代码说明
client
// 建立跟server的连接conn, err := grpc.Dial(love_const.Address, grpc.WithInsecure())
(左滑可查看完整代码,下同)
server
// 设置监听协议与端口lis, err := net.Listen("tcp", love_const.Address)// 初始化gRPC server实例grpcServer := grpc.NewServer()// 注册命令字与处理函数love_proto.RegisterBehaviorServer(grpcServer, newServer())// 启动服务grpcServer.Serve(lis)
0x1: A simple RPC
Selina职场工作不顺,刚被leader说了一顿,又恰好来大姨妈了,心情十分不好,问John:在哪
John正在打游戏,非常忙,未察觉出Selina的语气有异样,说:刚起床,在打游戏呢
Selina: "在哪"Jhon: "刚起床,在打游戏呢"
这种模式就是我们用得最多的模式,一发一收
client代码
response, err := client.WhereAreYou(ctx, message)
server代码
func (s *loveServer) WhereAreYou(ctx context.Context, message *love_proto.Message) (*love_proto.Response, error) {rsp := new(love_proto.Response)rsp.Words = "刚起床,在打游戏呢"return rsp, nil}
0x2:A client-to-server streaming RPC
Selina开始向John诉苦,说了很多话
John忙着打游戏,只看清“我来大姨妈了”,就只回了一句:哦,多喝热水
Selina: "你在干嘛"Selina: "我不开心"Selina: "我要和你说话"Selina: "你怎么还不打电话过来"Selina: "我来大姨妈了"Jhon: "哦,多喝热水"
这种模式是client先建立长连接,然后发送多个request给server,server最后统一回一个rsp。
应用场景
agent收集CPU,内存等数据到server
客户端心跳
客户端并发调用细小粒度的接口。比如有5个后台接口A B C D E,客户端在不同页面,可以调用不同的接口组合,比如在个人页,就调用ABC;在动态页面,就调用CDE,后台都只会有一个rsp。这种模式的好处就是让后台可以将接口的粒度细化,客户端调用灵活,减少重复代码,提高复用率
client代码
// 获取一个stream对象stream, err := client.ContinuousCall(ctx)if err != nil {log.Fatalf("%v.ContinuousCall(_) = _, %v", client, err)}// 通过stream来发送多个messagefor _, message := range messages {fmt.Printf("message words: %s\n", message.Words)if err := stream.Send(message); err != nil {log.Fatalf("%v.Send(%v) = %v", stream, message, err)}}// 发送EOF给server,然后收取server的回包reply, err := stream.CloseAndRecv()if err != nil {log.Fatalf("%v.CloseAndRecv() got error %v, want %v", stream, err, nil)}
server代码
for {// 不断收取client的发包message, err := stream.Recv()// 当客户端发送EOF时说明要给回包了if err == io.EOF {rsp := new(love_proto.Response)rsp.Words = "哦,多喝热水"return stream.SendAndClose(rsp)}if err != nil {return err}printGirlWords(message.Words)}
0x3: A server-to-client streaming RPC
Selina生气了
John先一直想解决方案,然后发现没有回应,就要买礼物,安慰等措施:包治百病
Selina: "这么久不给我打电话,你是不是不爱我了?"Jhon: "啊,宝贝你怎么了?"Jhon: "我刚刚在玩游戏,那一局刚开,我走不开啊"Jhon: "你不能不讲道理啊"Jhon: "你都20分钟不回我了"Jhon: "好了,宝贝,我错了,都是我的错"Jhon: "你在家等我,我过去接你,带你去买包包"
这种模式是client先发一个请求到server,然后server不停的回包
应用场景
股票app。客户端向服务端发送一个股票代码,服务端就把该股票的实时数据源源不断的返回给客户端
app的在线push。client先发请求到server注册,然后server就可以发在线push了
client代码
// 获取stream对象,并发送一个消息stream, err := client.LoveOrNot(ctx, message)if err != nil {log.Fatalf("%v.LoveOrNot(_) = _, %v", client, err)}for {// 不断收取server回包response, err := stream.Recv()// EOF表示server发包结束if err == io.EOF {break}if err != nil {log.Fatalf("%v.LoveOrNot(_) = _, %v", client, err)}log.Printf("rsp: %s", response.Words)}
server代码
for _, response := range responses {// 不停向client发消息if err := stream.Send(response); err != nil {return err}}
0x4: A Bidirectional streaming RPC
John道歉,买礼物,Selina不再生气,于是开始聊天
Selina: "好呀好呀"Jhon: "宝贝我很快就到啦"Selina: "等我化妆"Jhon: "宝贝我很快就到啦"Selina: "亲爱的你最好啦"Jhon: "宝贝我很快就到啦"Selina: "么么哒"Jhon: "宝贝我很快就到啦"
这种模式就是建立一个长连接,然后client和server可以随意收发数据
应用场景
聊天机器人
有状态的游戏服务器进行数据交换。比如LOL,王者荣耀等竞技游戏,client和server之间需要非常频繁地交换数据
client代码
// 获取stream对象stream, err := client.LoveChat(ctx)if err != nil {log.Fatalf("%v.LoveChat(_) = _, %v", client, err)}// 这个管道没有太多实际意义,只是为了让client将rsp都收完整waitc := make(chan struct{})go func() {for {// 不停地收server的包response, err := stream.Recv()// server回包EOF,然后关闭管道if err == io.EOF {// read done.close(waitc)log.Printf("EOF return")return}if err != nil {log.Fatalf("Failed to receive a note : %v", err)}log.Printf("rsp: %s", response.Words)}}()// 每隔一秒向server发一个消息,模拟聊天场景for _, message := range messages {time.Sleep(1 * time.Second)if err := stream.Send(message); err != nil {log.Fatalf("Failed to send a message: %v", err)}}// 这里告诉server: client已经将数据发完了(其实就是给server发一个EOF),server会返回一个io.EOFstream.CloseSend()
server代码
for {// 不断收取client发包message, err := stream.Recv()if err == io.EOF {return nil}if err != nil {return err}for _, response := range responses {// 给client发包if err := stream.Send(response); err != nil {return err}}}
男生们看过来,看过来
既然标题写了“用gRPC教程序员谈恋爱”,那么肯定就得给点干货。女生可能天生比男生更感性一些,所以当女生在倾诉问题的时候,千万不要只去想解决方案,而是要尽快地安慰女生,倾听她的吐槽,顺着她的思路去稍微吐槽下。可以换位思考,如果你自己心情不好的时候,肯定也想有人来安慰,而不是想听到”你要看开一点,这样工作效率才会高”之类的解决方案的话
在倾听和安慰之后,然后把问题解决,或者帮助女生把问题解决。做和说同样重要。
Attributes:
cardinality: A cardinality.Cardinality value.
style: A style.Service value.
unary_unary_inline: The implementation of the method as a callable value
that takes a request value and a ServicerContext object and returns a
response value. Only non-None if cardinality is
cardinality.Cardinality.UNARY_UNARY and style is style.Service.INLINE.
unary_stream_inline: The implementation of the method as a callable value
that takes a request value and a ServicerContext object and returns an
iterator of response values. Only non-None if cardinality is
cardinality.Cardinality.UNARY_STREAM and style is style.Service.INLINE.
stream_unary_inline: The implementation of the method as a callable value
that takes an iterator of request values and a ServicerContext object and
returns a response value. Only non-None if cardinality is
cardinality.Cardinality.STREAM_UNARY and style is style.Service.INLINE.
stream_stream_inline: The implementation of the method as a callable value
that takes an iterator of request values and a ServicerContext object and
returns an iterator of response values. Only non-None if cardinality is
cardinality.Cardinality.STREAM_STREAM and style is style.Service.INLINE.
unary_unary_event: The implementation of the method as a callable value that
takes a request value, a response callback to which to pass the response
value of the RPC, and a ServicerContext. Only non-None if cardinality is
cardinality.Cardinality.UNARY_UNARY and style is style.Service.EVENT.
unary_stream_event: The implementation of the method as a callable value
that takes a request value, a stream.Consumer to which to pass the
response values of the RPC, and a ServicerContext. Only non-None if
cardinality is cardinality.Cardinality.UNARY_STREAM and style is
style.Service.EVENT.
stream_unary_event: The implementation of the method as a callable value
that takes a response callback to which to pass the response value of the
RPC and a ServicerContext and returns a stream.Consumer to which the
request values of the RPC should be passed. Only non-None if cardinality
is cardinality.Cardinality.STREAM_UNARY and style is style.Service.EVENT.
stream_stream_event: The implementation of the method as a callable value
that takes a stream.Consumer to which to pass the response values of the
RPC and a ServicerContext and returns a stream.Consumer to which the
request values of the RPC should be passed. Only non-None if cardinality
is cardinality.Cardinality.STREAM_STREAM and style is
style.Service.EVENT.