「推荐系统从0到1」服务发现
Posted 算法工厂
tags:
篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了「推荐系统从0到1」服务发现相关的知识，希望对你有一定的参考价值。
前沿
首先谈谈我对推荐系统的引擎和算法的理解。
现在市面上讲起推荐系统，大多都是讲各种算法，讲的天花乱坠，高深莫测，其实很多算法都是大同小异，核心思想是差不多的，只不过实现手段略有差异。而在工业上，各种复杂算法能够落地的，我认为不多，大部分的厂商，运用的算法都是很集中的那一部分算法。
一套好的推荐系统，对于引擎是非常依赖的，实验显示，响应时长与各项指标之间都是有直接关联的，响应时长越长，指标越低。
作为一个朴实的推荐码农，我还是想从基础做起，朴朴实实，脚踏实地，先把引擎部分做好。当然，算法后面也会有，毕竟引擎和算法缺一不可。
那么，废话少说，推荐引擎，搞起吧！
服务发现
既然是搞引擎，也就是后端，当然要先把架构先搭建起来。
后端服务，微服务已经成为了当前的主流，具有非常多的优点，比如高内聚，单独部署，各自负载均衡等，当然缺点也有，通信更复杂了等。具体就不在这里展开了，有兴趣的兄弟们可以百度，google一下。
在整个流程中，可以分为服务端的要做的，以及客户端要做的，下面依次来看一下。
服务端比较简单，只需要将自己的信息存储到某个存储中。
是不是原理上非常简单！那么进入实操吧！
etcd介绍
以前有zookeeper，而zookeeper可以看到，早就不再维护更新了。
而etcd，用go语言开发，因kubernetes而闻名，在kubernetes中，使用etcd作为分布式存储获取分布式锁。
所以我们当然用更年轻，更轻量，并且也非常稳定的etcd搞了！就是这么喜新厌旧= =
etcd使用raft算法实现的一致性，至于raft算法，可以看下面这个动画演示，很完美生动。
raft动画演示
etcd实战
我这边用docker来做自己的测试环境，上我的docker-compose.yaml
version: '2.2'
services:
etcd:
image: gcr.io/etcd-development/etcd:v3.4.13
container_name: etcd
restart: always
ports:
- 2379:2379
- 2380:2380
command:
- "/usr/local/bin/etcd"
- "--name"
- "s1"
- "--data-dir"
- "/etcd-data"
- "--advertise-client-urls"
- "http://0.0.0.0:2379"
- "--listen-client-urls"
- "http://0.0.0.0:2379"
- "--initial-advertise-peer-urls"
- "http://0.0.0.0:2380"
- "--listen-peer-urls"
- "http://0.0.0.0:2380"
- "--initial-cluster-token"
- "tkn"
- "--initial-cluster"
- "s1=http://0.0.0.0:2380"
- "--initial-cluster-state"
- "new"
如果想通过其他途径安装可以看官方的说明：
安装etcd
那么，既然是存储，我们就来测试一下CRUD吧，还有etcd的租约功能。
CRUD:
# etcdctl put test/key hello
OK
# etcdctl get test/key
test/key
hello
# etcdctl put test/key goodbye
OK
# etcdctl get test/key
test/key
goodbye
# etcdctl del test/key
1
# etcdctl get test/key
租约：
创建租约，120s过期
# etcdctl lease grant 120
lease 3f3575c45fa5ff26 granted with TTL(120s)
查看租约列表
# etcdctl lease list
found 1 leases
3f3575c45fa5ff26
新建kv，并绑定租约
# etcdctl put test/key hello --lease="3f3575c45fa5ff26"
OK
查看租约下的key剩余时间
# etcdctl lease timetolive 3f3575c45fa5ff26 --keys
lease 3f3575c45fa5ff26 granted with TTL(120s), remaining(46s), attached keys([test/key])
查看还存在的key
# etcdctl get --prefix ""
test/key
hello
等租约过期后，查看key，key已被自动删除
# etcdctl lease timetolive 3f3575c45fa5ff26 --keys
lease 3f3575c45fa5ff26 already expired
# etcdctl get --prefix ""
租约续约：
同样创建租约，绑定kv
# etcdctl lease grant 30
lease 3f3575c45fa5ff2c granted with TTL(30s)
# etcdctl put test/key hello --lease="3f3575c45fa5ff2c"
OK
续约
# etcdctl lease keep-alive 3f3575c45fa5ff2c
lease 3f3575c45fa5ff2c keepalived with TTL(30)
lease 3f3575c45fa5ff2c keepalived with TTL(30)
lease 3f3575c45fa5ff2c keepalived with TTL(30)
lease 3f3575c45fa5ff2c keepalived with TTL(30)
打开个新窗口查看租约与key
# etcdctl lease timetolive 3f3575c45fa5ff2c --keys
lease 3f3575c45fa5ff2c granted with TTL(30s), remaining(23s), attached keys([test/key])
# etcdctl get --prefix ""
test/key
hello
发现并没有过期。
golang+grpc+etcd 服务发现终极实战！
先上github仓库：https://github.com/feng-jin/recommender-system
代码目录/go_server/src/lib/discovery/
说一下整个流程：
服务端向etcd注册服务，就是将本服务的信息写进etcd。
客户端大体流程：
-
从etcd取服务端地址列表，并watch列表变化，并更新。 -
把地址列表写进grpc resolver的resolver.ClientConn的地址列表中。 -
grpc创建连接，根据负载均衡请求。
整个模块分为7个文件：
-
config.go，配置文件。 -
discovery.go，用于初始化。 -
register.go，用于服务注册。 -
resolver.go，用于解析etcd里注册的服务地址，以及grpc负载均衡。 -
util.go，公共方法。 -
wrapper.go，对外部提供的调用封装。 -
ctx.go，context，设置超时时间。
config.go
package config
import "time"
// etcd
const (
Timeout = 15 * time.Second
Expires = 10
TickerInterval = 5
// scheme
Scheme = "etcd"
// etcd中存储key的格式前缀：/scheme/authority/endpoint
DirFormat = "/%s/%s/%s/"
// grpc resolver中自定义解析需要提供的格式：scheme://authority/endpoint
// 其中scheme可以理解为解析策略，authority可以理解为权限管理，endpoint为地址
TargetFormat = "%s://%s/%s"
)
// server name
const (
GreetServer = "greet_server"
)
discovery.go
package discovery
import (
"fmt"
"go_server/src/lib/discovery/config"
"go_server/src/lib/logger"
"strings"
"go.etcd.io/etcd/clientv3"
)
var (
client *clientv3.Client
)
// Init 初始化etcd
func Init(etcdAddr string) error {
var err error
if client == nil {
//构建etcd client
client, err = clientv3.New(clientv3.Config{
Endpoints: strings.Split(etcdAddr, ";"),
DialTimeout: config.Timeout,
})
if err != nil {
logger.Error("连接etcd失败：%s\n", err)
fmt.Printf("连接etcd失败：%s\n", err)
return err
}
}
return nil
}
register.go
package discovery
import (
"context"
"errors"
"fmt"
"go_server/src/lib/discovery/config"
"os"
"os/signal"
"syscall"
"time"
"go.etcd.io/etcd/clientv3"
)
//Service 服务端用于服务注册的对象
type Service struct {
Name string //服务名称
Host string //{ip}:{port}
Env string //所属环境
Key string //保存在etcd中的key
}
var service *Service
func (s *Service) register() error {
if s.Env == "" {
return errors.New("env is null")
}
s.Key = fmt.Sprintf(config.DirFormat, config.Scheme, s.Env, s.Name) + s.Host
ticker := time.NewTicker(time.Second * time.Duration(config.TickerInterval))
go func() {
for {
resp, err := client.Get(context.Background(), s.Key)
if err != nil {
fmt.Printf("获取服务地址失败：%s", err)
} else if resp.Count == 0 { //尚未注册
err = s.keepAlive()
if err != nil {
fmt.Printf("保持连接失败：%s", err)
}
}
<-ticker.C
}
}()
return nil
}
// keepAlive 创建租约，绑定，并续期
func (s *Service) keepAlive() error {
//创建租约
leaseResp, err := client.Grant(context.Background(), config.Expires)
if err != nil {
fmt.Printf("创建租期失败：%s\n", err)
return err
}
//将服务地址注册到etcd中
_, err = client.Put(context.Background(), s.Key, s.Host, clientv3.WithLease(leaseResp.ID))
if err != nil {
fmt.Printf("注册服务失败：%s", err)
return err
}
//租约续期
ch, err := client.KeepAlive(context.Background(), leaseResp.ID)
if err != nil {
fmt.Printf("租约续期失败：%s\n", err)
return err
}
//清空keepAlive返回的channel
go func() {
for {
<-ch
}
}()
return nil
}
//取消注册
func (s *Service) unRegister() {
if client != nil {
_, _ = client.Delete(context.Background(), s.Key)
}
}
func WaitForClose() {
ch := make(chan os.Signal, 1)
signal.Notify(ch, syscall.SIGTERM, syscall.SIGINT, syscall.SIGKILL, syscall.SIGHUP, syscall.SIGQUIT)
sig := <-ch
service.unRegister()
if i, ok := sig.(syscall.Signal); ok {
os.Exit(int(i))
} else {
os.Exit(0)
}
}
resolver.go
package discovery
import (
"context"
"fmt"
"go_server/src/lib/discovery/config"
"strings"
"go.etcd.io/etcd/clientv3"
"google.golang.org/grpc"
"google.golang.org/grpc/resolver"
)
//EtcdResolver解析器
type EtcdResolver struct {
dir string
clientConn resolver.ClientConn
}
func Resolver(env string, name string) *grpc.ClientConn {
//注册etcd解析器
r := &EtcdResolver{}
resolver.Register(r)
target := fmt.Sprintf(config.TargetFormat, r.Scheme(), env, name)
//客户端连接服务器(负载均衡：轮询) 会同步调用r.Build()
dailOpts := []grpc.DialOption{
grpc.WithBalancerName("round_robin"), // grpc内部提供的轮询负载均衡
grpc.WithInsecure(),
grpc.WithDefaultCallOptions(
grpc.MaxCallRecvMsgSize(1024 * 1024 * 16),
),
}
conn, err := grpc.Dial(target, dailOpts...)
if err != nil {
fmt.Println("连接服务器失败：", err)
}
return conn
}
func (r *EtcdResolver) Scheme() string {
return config.Scheme
}
//构建解析器 grpc.Dial()同步调用
func (r *EtcdResolver) Build(target resolver.Target, clientConn resolver.ClientConn, _ resolver.BuildOptions) (resolver.Resolver, error) {
r.clientConn = clientConn
r.dir = fmt.Sprintf(config.DirFormat, target.Scheme, target.Authority, target.Endpoint)
go r.watch()
return r, nil
}
//监听etcd中某个key前缀的服务地址列表的变化
func (r *EtcdResolver) watch() {
//初始化服务地址列表
var addrList []resolver.Address
resp, err := client.Get(context.Background(), r.dir, clientv3.WithPrefix())
if err != nil {
fmt.Println("获取服务地址列表失败：", err)
} else {
for i := range resp.Kvs {
fmt.Println(strings.TrimPrefix(string(resp.Kvs[i].Key), r.dir))
addrList = append(addrList, resolver.Address{Addr: strings.TrimPrefix(string(resp.Kvs[i].Key), r.dir)})
}
}
r.clientConn.NewAddress(addrList)
//监听服务地址列表的变化
rch := client.Watch(context.Background(), r.dir, clientv3.WithPrefix())
for n := range rch {
for _, ev := range n.Events {
addr := strings.TrimPrefix(string(ev.Kv.Key), r.dir)
switch ev.Type {
case clientv3.EventTypePut:
if !exists(addrList, addr) {
addrList = append(addrList, resolver.Address{Addr: addr})
r.clientConn.NewAddress(addrList)
}
case clientv3.EventTypeDelete:
if s, ok := remove(addrList, addr); ok {
addrList = s
r.clientConn.NewAddress(addrList)
}
}
}
}
}
func exists(l []resolver.Address, addr string) bool {
for i := range l {
if l[i].Addr == addr {
return true
}
}
return false
}
func remove(s []resolver.Address, addr string) ([]resolver.Address, bool) {
for i := range s {
if s[i].Addr == addr {
s[i] = s[len(s)-1]
return s[:len(s)-1], true
}
}
return nil, false
}
//Close ...
func (r *EtcdResolver) Close() {}
//ResolveNow ...
func (r *EtcdResolver) ResolveNow(_ resolver.ResolveNowOption) {}
util.go
package discovery
import (
"fmt"
"net"
)
// 获取本机ip地址
func getIntranetIP() (ip string) {
if addrs, err := net.InterfaceAddrs(); err == nil {
for _, address := range addrs {
if ipnet, ok := address.(*net.IPNet); ok && !ipnet.IP.IsLoopback() {
if ipnet.IP.To4() != nil {
ip = ipnet.IP.String()
break
}
}
}
}
return
}
// 自动获取本机的ip以及端口号，ip:port格式
func getListener() (listener net.Listener, host string, err error) {
host = "0.0.0.0:0"
listener, err = net.Listen("tcp", host)
if err == nil {
addr := listener.Addr().String()
_, portString, _ := net.SplitHostPort(addr)
host = fmt.Sprintf("%s:%s", getIntranetIP(), portString)
}
return
}
wrapper.go
package discovery
import (
"fmt"
"go_server/src/lib/discovery/config"
"go_server/src/lib/proto/greet"
"google.golang.org/grpc"
)
func GreetRegister(env string, server greet.GreetServer) error {
listener, host, err := getListener()
if err != nil {
fmt.Println("监听网络失败：", err)
return err
}
fmt.Println("host:", host)
srv := grpc.NewServer()
go srv.Serve(listener)
greet.RegisterGreetServer(srv, server)
service = &Service{Name: config.GreetServer, Host: host, Env: env}
err = service.register()
if err != nil {
fmt.Println(err)
return err
}
return nil
}
func GreetResolve(env string) greet.GreetClient {
return greet.NewGreetClient(Resolver(env, config.GreetServer))
}
ctx.go
package discovery
import (
"context"
"time"
)
// 1s超时
func Context1s() (ctx context.Context, cancel context.CancelFunc) {
return context.WithTimeout(context.TODO(), time.Second)
}
测试一下吧，测试文件也都在github仓库里：
搞个测试的proto，server和client，也直接上代码：
greet.proto
syntax = "proto3";
option go_package = "src/lib/proto/greet";
service Greet {
rpc Morning(GreetRequest)returns(GreetResponse){}
rpc Night(GreetRequest)returns(GreetResponse){}
}
message GreetRequest {
string name = 1;
}
message GreetResponse {
string message = 1;
string from = 2;
}
server main.go
package main
import (
"context"
"flag"
"fmt"
"go_server/src/lib/discovery"
proto "go_server/src/lib/proto/greet"
)
var (
Flag = flag.String("flag", "a", "flag")
EtcdAddr = flag.String("EtcdAddr", "127.0.0.1:2379", "register etcd address")
Env = flag.String("Env", "test", "env")
)
//rpc服务接口
type GreetServer struct{}
func (gs *GreetServer) Morning(ctx context.Context, req *proto.GreetRequest) (*proto.GreetResponse, error) {
fmt.Printf("Morning 调用: %s\n", req.Name)
return &proto.GreetResponse{
Message: "Good morning, " + req.Name,
From: *Flag,
}, nil
}
func (gs *GreetServer) Night(ctx context.Context, req *proto.GreetRequest) (*proto.GreetResponse, error) {
fmt.Printf("Night 调用: %s\n", req.Name)
return &proto.GreetResponse{
Message: "Good night, " + req.Name,
From: *Flag,
}, nil
}
func main() {
flag.Parse()
err := discovery.Init(*EtcdAddr)
if err != nil {
fmt.Println(err)
return
}
err = discovery.GreetRegister(*Env, &GreetServer{})
if err != nil {
fmt.Println(err)
return
}
discovery.WaitForClose()
}
client main.go
package main
import (
"flag"
"fmt"
"go_server/src/lib/discovery"
proto "go_server/src/lib/proto/greet"
"time"
)
var (
EtcdAddr = flag.String("EtcdAddr", "127.0.0.1:2379", "register etcd address")
Env = flag.String("Env", "test", "env")
)
func main() {
flag.Parse()
err := discovery.Init(*EtcdAddr)
if err != nil {
fmt.Println(err)
return
}
c := discovery.GreetResolve(*Env)
ticker := time.NewTicker(1 * time.Second)
for range ticker.C {
fmt.Println("Morning 调用...")
ctx, cancel := discovery.Context1s()
resp1, err := c.Morning(
ctx,
&proto.GreetRequest{Name: "Jinfeng"},
)
cancel()
if err != nil {
fmt.Println("Morning调用失败：", err)
return
}
fmt.Printf("Morning 响应：%s，来自：%s\n", resp1.Message, resp1.From)
fmt.Println("Night 调用...")
ctx, cancel = discovery.Context1s()
resp2, err := c.Night(
ctx,
&proto.GreetRequest{Name: "Jinfeng"},
)
cancel()
if err != nil {
fmt.Println("Night调用失败：", err)
return
}
fmt.Printf("Night 响应：%s，来自：%s\n", resp2.Message, resp2.From)
}
}
跑起来吧，起3个server，可以看到，在etcd已经注册了3台服务。
# etcdctl get --prefix ""
/etcd/test/greet_server/192.168.31.71:52963
192.168.31.71:52963
/etcd/test/greet_server/192.168.31.71:52969
192.168.31.71:52969
/etcd/test/greet_server/192.168.31.71:52973
192.168.31.71:52973
client调用
➜ client git:(main) ✗ go run .
192.168.31.71:52963
192.168.31.71:52969
192.168.31.71:52973
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：c
Night 调用...
Night 响应：Good night, Jinfeng，来自：a
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：b
Night 调用...
Night 响应：Good night, Jinfeng，来自：c
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：a
Night 调用...
Night 响应：Good night, Jinfeng，来自：b
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：c
shutdown一台服务
Morning 响应：Good morning, Jinfeng，来自：a
Night 调用...
Night 响应：Good night, Jinfeng，来自：b
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：a
Night 调用...
Night 响应：Good night, Jinfeng，来自：b
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：a
Night 调用...
Night 响应：Good night, Jinfeng，来自：b
重新启动
Morning 响应：Good morning, Jinfeng，来自：a
Night 调用...
Night 响应：Good night, Jinfeng，来自：b
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：c
Night 调用...
Night 响应：Good night, Jinfeng，来自：a
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：b
Night 调用...
Night 响应：Good night, Jinfeng，来自：c
Morning 调用...
Morning 响应：Good morning, Jinfeng，来自：a
Night 调用...
Night 响应：Good night, Jinfeng，来自：b
这一轮，只是用grpc内部简单的轮训来做负载均衡，后面有空了，再加入一致性哈希等方法吧！
到现在，服务发现已经有了，下面就可以先做一个简单的推荐系统，把流程跑起来了！
后面计划先做一个只有简单召回的推荐系统，然后再慢慢优化整套系统。
兄弟们，奥利给！
以上是关于「推荐系统从0到1」服务发现的主要内容，如果未能解决你的问题，请参考以下文章