redis缓存替换+pgvector向量替换

ai-chat-backend/services/ai-chat-service/ai-chat-service.go

@@ -0,0 +1,23 @@
+package ai_chat_service
+
+import (
+	"ai-chat-backend/pkg/config"
+	grpc_client "ai-chat-backend/services/grpc-client"
+	"sync"
+)
+
+var pool grpc_client.ClientPool
+var once sync.Once
+
+type client struct {
+	grpc_client.DefaultClient
+}
+
+func GetAiChatServiceClientPool() grpc_client.ClientPool {
+	once.Do(func() {
+		cnf := config.GetConfig()
+		c := &client{}
+		pool = c.GetPool(cnf.DependOn.AiChatService.Address)
+	})
+	return pool
+}

ai-chat-backend/services/ai-chat-service/proto/chat.proto

@@ -0,0 +1,72 @@
+syntax = "proto3";
+option go_package = "ai-chat-service/proto";
+package ai_chat_service.zvoice.com;
+
+message ChatCompletionRequest {
+  string message = 1[json_name = "message"];
+  string id = 2[json_name = "id"];
+  string pid = 3[json_name = "p_id"];
+  bool enableContext = 4[json_name = "enable_context"];
+  ChatParam chatParam = 5[json_name = "chat_param"];
+}
+
+message ChatParam {
+  string model = 1[json_name = "model"];
+  int32  maxTokens = 2[json_name = "max_tokens"];
+  float temperature = 3[json_name = "temperature"];
+  float topP = 4[json_name = "top_p"];
+  float presencePenalty = 5[json_name = "presence_penalty"];
+  float frequencyPenalty = 6[json_name = "frequency_penalty"];
+  string botDesc =7[json_name = "bot_desc"];
+  int32 minResponseTokens = 8[json_name = "min_response_tokens"];
+  int32 contextTTL = 9[json_name = "context_ttl"];
+  int32 contextLen = 10[json_name = "context_len"];
+}
+
+// 服务响应消息，非流式响应
+message ChatCompletionResponse {
+  string id = 1 [json_name = "id"];
+  string object = 2 [json_name = "object"];
+  int64 created = 3 [json_name = "created"];
+  string model = 4 [json_name = "model"];
+  repeated  ChatCompletionChoice choices = 5 [json_name = "choices"];
+  Usage usage = 6[json_name = "usage"];
+}
+message ChatCompletionChoice {
+  int32 index = 1[json_name = "index"];
+  ChatCompletionMessage message = 2 [json_name = "message"];
+  string finishReason = 3[json_name = "finish_reason"];
+}
+message ChatCompletionMessage {
+  string role=1[json_name = "role"];
+  string content=2[json_name = "content"];
+  string name=3[json_name = "name"];
+}
+message Usage {
+  int32 promptTokens = 1 [json_name = "prompt_tokens"];
+  int32 completionTokens =2 [json_name = "completion_tokens"];
+  int32 totalTokens      = 3 [json_name = "total_tokens"];
+}
+
+// 服务响应消息，流式响应
+message  ChatCompletionStreamResponse {
+  string id = 1 [json_name = "id"];
+  string object = 2 [json_name = "object"];
+  int64 created = 3 [json_name = "created"];
+  string model = 4 [json_name = "model"];
+  repeated  ChatCompletionStreamChoice choices = 5 [json_name = "choices"];
+}
+message ChatCompletionStreamChoice {
+  int32 index = 1[json_name = "index"];
+  ChatCompletionStreamChoiceDelta delta =2 [json_name = "delta"];
+  string finishReason = 3[json_name="finish_reason"];
+}
+message ChatCompletionStreamChoiceDelta {
+  string content = 1 [json_name = "content"];
+  string role = 2 [json_name = "role"];
+}
+
+service Chat {
+  rpc ChatCompletion(ChatCompletionRequest) returns (ChatCompletionResponse);
+  rpc ChatCompletionStream(ChatCompletionRequest) returns (stream ChatCompletionStreamResponse);
+}

ai-chat-backend/services/ai-chat-service/proto/chat_grpc.pb.go

@@ -0,0 +1,169 @@
+// Code generated by protoc-gen-go-grpc. DO NOT EDIT.
+// versions:
+// - protoc-gen-go-grpc v1.2.0
+// - protoc             v4.22.0
+// source: proto/chat.proto
+
+package proto
+
+import (
+	context "context"
+	grpc "google.golang.org/grpc"
+	codes "google.golang.org/grpc/codes"
+	status "google.golang.org/grpc/status"
+)
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the grpc package it is being compiled against.
+// Requires gRPC-Go v1.32.0 or later.
+const _ = grpc.SupportPackageIsVersion7
+
+// ChatClient is the client API for Chat service.
+//
+// For semantics around ctx use and closing/ending streaming RPCs, please refer to https://pkg.go.dev/google.golang.org/grpc/?tab=doc#ClientConn.NewStream.
+type ChatClient interface {
+	ChatCompletion(ctx context.Context, in *ChatCompletionRequest, opts ...grpc.CallOption) (*ChatCompletionResponse, error)
+	ChatCompletionStream(ctx context.Context, in *ChatCompletionRequest, opts ...grpc.CallOption) (Chat_ChatCompletionStreamClient, error)
+}
+
+type chatClient struct {
+	cc grpc.ClientConnInterface
+}
+
+func NewChatClient(cc grpc.ClientConnInterface) ChatClient {
+	return &chatClient{cc}
+}
+
+func (c *chatClient) ChatCompletion(ctx context.Context, in *ChatCompletionRequest, opts ...grpc.CallOption) (*ChatCompletionResponse, error) {
+	out := new(ChatCompletionResponse)
+	err := c.cc.Invoke(ctx, "/ai_chat_service.zvoice.com.Chat/ChatCompletion", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *chatClient) ChatCompletionStream(ctx context.Context, in *ChatCompletionRequest, opts ...grpc.CallOption) (Chat_ChatCompletionStreamClient, error) {
+	stream, err := c.cc.NewStream(ctx, &Chat_ServiceDesc.Streams[0], "/ai_chat_service.zvoice.com.Chat/ChatCompletionStream", opts...)
+	if err != nil {
+		return nil, err
+	}
+	x := &chatChatCompletionStreamClient{stream}
+	if err := x.ClientStream.SendMsg(in); err != nil {
+		return nil, err
+	}
+	if err := x.ClientStream.CloseSend(); err != nil {
+		return nil, err
+	}
+	return x, nil
+}
+
+type Chat_ChatCompletionStreamClient interface {
+	Recv() (*ChatCompletionStreamResponse, error)
+	grpc.ClientStream
+}
+
+type chatChatCompletionStreamClient struct {
+	grpc.ClientStream
+}
+
+func (x *chatChatCompletionStreamClient) Recv() (*ChatCompletionStreamResponse, error) {
+	m := new(ChatCompletionStreamResponse)
+	if err := x.ClientStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
+// ChatServer is the server API for Chat service.
+// All implementations must embed UnimplementedChatServer
+// for forward compatibility
+type ChatServer interface {
+	ChatCompletion(context.Context, *ChatCompletionRequest) (*ChatCompletionResponse, error)
+	ChatCompletionStream(*ChatCompletionRequest, Chat_ChatCompletionStreamServer) error
+	mustEmbedUnimplementedChatServer()
+}
+
+// UnimplementedChatServer must be embedded to have forward compatible implementations.
+type UnimplementedChatServer struct {
+}
+
+func (UnimplementedChatServer) ChatCompletion(context.Context, *ChatCompletionRequest) (*ChatCompletionResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method ChatCompletion not implemented")
+}
+func (UnimplementedChatServer) ChatCompletionStream(*ChatCompletionRequest, Chat_ChatCompletionStreamServer) error {
+	return status.Errorf(codes.Unimplemented, "method ChatCompletionStream not implemented")
+}
+func (UnimplementedChatServer) mustEmbedUnimplementedChatServer() {}
+
+// UnsafeChatServer may be embedded to opt out of forward compatibility for this service.
+// Use of this interface is not recommended, as added methods to ChatServer will
+// result in compilation errors.
+type UnsafeChatServer interface {
+	mustEmbedUnimplementedChatServer()
+}
+
+func RegisterChatServer(s grpc.ServiceRegistrar, srv ChatServer) {
+	s.RegisterService(&Chat_ServiceDesc, srv)
+}
+
+func _Chat_ChatCompletion_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ChatCompletionRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(ChatServer).ChatCompletion(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/ai_chat_service.zvoice.com.Chat/ChatCompletion",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(ChatServer).ChatCompletion(ctx, req.(*ChatCompletionRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Chat_ChatCompletionStream_Handler(srv interface{}, stream grpc.ServerStream) error {
+	m := new(ChatCompletionRequest)
+	if err := stream.RecvMsg(m); err != nil {
+		return err
+	}
+	return srv.(ChatServer).ChatCompletionStream(m, &chatChatCompletionStreamServer{stream})
+}
+
+type Chat_ChatCompletionStreamServer interface {
+	Send(*ChatCompletionStreamResponse) error
+	grpc.ServerStream
+}
+
+type chatChatCompletionStreamServer struct {
+	grpc.ServerStream
+}
+
+func (x *chatChatCompletionStreamServer) Send(m *ChatCompletionStreamResponse) error {
+	return x.ServerStream.SendMsg(m)
+}
+
+// Chat_ServiceDesc is the grpc.ServiceDesc for Chat service.
+// It's only intended for direct use with grpc.RegisterService,
+// and not to be introspected or modified (even as a copy)
+var Chat_ServiceDesc = grpc.ServiceDesc{
+	ServiceName: "ai_chat_service.zvoice.com.Chat",
+	HandlerType: (*ChatServer)(nil),
+	Methods: []grpc.MethodDesc{
+		{
+			MethodName: "ChatCompletion",
+			Handler:    _Chat_ChatCompletion_Handler,
+		},
+	},
+	Streams: []grpc.StreamDesc{
+		{
+			StreamName:    "ChatCompletionStream",
+			Handler:       _Chat_ChatCompletionStream_Handler,
+			ServerStreams: true,
+		},
+	},
+	Metadata: "proto/chat.proto",
+}

ai-chat-backend/services/grpc-client/default.go

@@ -0,0 +1,28 @@
+package grpc_client
+
+import (
+	"ai-chat-backend/pkg/log"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/credentials/insecure"
+)
+
+type ServiceClient interface {
+	GetPool(addr string) ClientPool
+}
+type DefaultClient struct {
+}
+
+func (c *DefaultClient) GetPool(addr string) ClientPool {
+	pool, err := NewPool(addr, c.getOptions()...)
+	if err != nil {
+		log.Error(err)
+		return nil
+	}
+	return pool
+}
+
+func (c *DefaultClient) getOptions() []grpc.DialOption {
+	opts := make([]grpc.DialOption, 0)
+	opts = append(opts, grpc.WithTransportCredentials(insecure.NewCredentials()))
+	return opts
+}

ai-chat-backend/services/grpc-client/grpc_client_pool.go

@@ -0,0 +1,48 @@
+package grpc_client
+
+import (
+	"ai-chat-backend/pkg/log"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/connectivity"
+	"sync"
+)
+
+type ClientPool interface {
+	Get() *grpc.ClientConn
+	Put(*grpc.ClientConn)
+}
+
+type clientPool struct {
+	pool sync.Pool
+}
+
+func NewPool(target string, opts ...grpc.DialOption) (ClientPool, error) {
+	return &clientPool{
+		pool: sync.Pool{
+			New: func() any {
+				conn, err := grpc.NewClient(target, opts...)
+				if err != nil {
+					log.Error(err)
+					return nil
+				}
+				return conn
+			},
+		},
+	}, nil
+}
+
+func (c *clientPool) Get() *grpc.ClientConn {
+	conn := c.pool.Get().(*grpc.ClientConn)
+	if conn.GetState() == connectivity.Shutdown || conn.GetState() == connectivity.TransientFailure {
+		conn.Close()
+		conn = c.pool.New().(*grpc.ClientConn)
+	}
+	return conn
+}
+func (c *clientPool) Put(conn *grpc.ClientConn) {
+	if conn.GetState() == connectivity.Shutdown || conn.GetState() == connectivity.TransientFailure {
+		conn.Close()
+		return
+	}
+	c.pool.Put(conn)
+}

ai-chat-backend/services/services.go

@@ -0,0 +1,11 @@
+package services
+
+import (
+	"context"
+	"google.golang.org/grpc/metadata"
+)
+
+func AppendBearerTokenToContext(ctx context.Context, accessToken string) context.Context {
+	md := metadata.Pairs("Authorization", "Bearer "+accessToken)
+	return metadata.NewOutgoingContext(ctx, md)
+}