在阅读 go-zero 源码之前我们先来看看常用的负载均衡算法,看看其原理,以及是如何实现,然后我们在用这些负载均衡算法来和 go-zero 的对比下,看看各自的优缺点是啥。
轮询
proxy 服务与 ndoe 服务配置文件
{
"proxy": {
"url": "127.0.0.1:8080"
},
"nodes": [
{
"url": "127.0.0.1:8081"
},
{
"url": "127.0.0.1:8082"
},
{
"url": "127.0.0.1:8083"
}
]
}
proxy 服务、 ndoe 服务、轮询算法代码
// 配置
type Config struct {
Proxy Proxy `json:"proxy"`
Nodes []*Node `json:"nodes"`
}
// proxy 服务器配置
type Proxy struct {
Url string `json:"url"`
}
// node 服务器配置
type Node struct {
URL string `json:"url"`
IsDead bool
useCount int
mu sync.RWMutex
}
var cfg Config
func init() {
// 加载配置文件
data, err := ioutil.ReadFile("./config.json")
if err != nil {
log.Fatal(err.Error())
}
json.Unmarshal(data, &cfg)
}
// 设置 node 服务器宕机状态
func (node *Node) SetDead(b bool) {
node.mu.Lock()
node.IsDead = b
node.mu.Unlock()
}
// 获取 node 服务器是否宕机
func (node *Node) GetIsDead() bool {
node.mu.RLock()
isAlive := node.IsDead
node.mu.RUnlock()
return isAlive
}
var (
mu sync.Mutex
idx int = 0
)
// 轮询算法
func rrlbbHandler(w http.ResponseWriter, r *http.Request) {
maxLen := len(cfg.Nodes)
// Round Robin
mu.Lock()
currentNode := cfg.Nodes[idx%maxLen] // 循环数组
if currentNode.GetIsDead() {
idx++ // 如果 node 宕机,则轮询到下一个 node
currentNode = cfg.Nodes[idx%maxLen]
}
currentNode.useCount++
targetURL, err := url.Parse("http://" + currentNode.URL)
log.Println(targetURL.Host)
if err != nil {
log.Fatal(err.Error())
}
idx++
mu.Unlock()
reverseProxy := httputil.NewSingleHostReverseProxy(targetURL)
reverseProxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, e error) {
log.Printf("%v is dead.", targetURL)
currentNode.SetDead(true)
rrlbbHandler(w, r) // 节点宕机 递归调用自己
}
reverseProxy.ServeHTTP(w, r)
}
// node是否存活
func isAlive(url *url.URL) bool {
conn, err := net.DialTimeout("tcp", url.Host, time.Minute*1)
if err != nil {
log.Printf("Unreachable to %v, error %s:", url.Host, err.Error())
return false
}
defer conn.Close()
return true
}
// node探活
func healthCheck() {
t := time.NewTicker(time.Minute * 1)
for {
select {
case <-t.C:
for _, node := range cfg.Nodes {
pingURL, err := url.Parse(node.URL)
if err != nil {
log.Fatal(err.Error())
}
isAlive := isAlive(pingURL)
node.SetDead(!isAlive)
msg := "ok"
if !isAlive {
msg = "dead"
}
log.Printf("%v checked %s by healthcheck", node.URL, msg)
}
}
}
}
// 启动 proxy 服务
func proxyServerStart() {
var err error
go healthCheck()
s := http.Server{
Addr: cfg.Proxy.Url,
Handler: http.HandlerFunc(rrlbbHandler),
}
if err = s.ListenAndServe(); err != nil {
log.Fatal(err.Error())
}
}
// 启动所有 node 服务
func nodeServerStart() {
http.HandleFunc("/ping", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte("pong"))
})
wg := new(sync.WaitGroup)
wg.Add(len(cfg.Nodes))
for i, node := range cfg.Nodes {
go func() {
if i > 0 {
// 模拟一个node宕机
log.Fatal(http.ListenAndServe(node.URL, nil))
}
wg.Done()
}()
time.Sleep(time.Millisecond * 100)
}
wg.Wait()
}
最核心的算法就是这一段,非常简单,轮询的本质其实是循环数组
currentNode := cfg.Nodes[idx%maxLen] // 数组循环
if currentNode.GetIsDead() {
idx++ // 如果 node 宕机,则轮询到下一个 node
currentNode = cfg.Nodes[idx%maxLen]
}
我们来编写测试代码来测试下吧
func Test_RoundRobinBalancer(t *testing.T) {
go nodeServerStart()
time.Sleep(time.Millisecond * 200)
go proxyServerStart()
time.Sleep(time.Millisecond * 100)
for _, tt := range [...]struct {
name, method, uri string
body io.Reader
want *http.Request
wantBody string
}{
{
name: "GET with ping url",
method: "GET",
uri: "http://127.0.0.1:8080/ping",
body: nil,
wantBody: "pong",
},
} {
t.Run(tt.name, func(t *testing.T) {
for i := 1; i <= 10; i++ {
body, err := utils.HttpRequest(tt.method, tt.uri, tt.body)
if err != nil {
t.Errorf("ReadAll: %v", err)
}
if string(body) != tt.wantBody {
t.Errorf("Body = %q; want %q", body, tt.wantBody)
}
}
for _, node := range cfg.Nodes {
log.Printf("node: %s useCount: %d", node.URL, node.useCount)
}
})
}
}
测试结果如下:
-------- node 调度顺序--------
2022/04/06 19:50:24 127.0.0.1:8081
2022/04/06 19:50:24 http://127.0.0.1:8081 is dead.
2022/04/06 19:50:24 127.0.0.1:8082
2022/04/06 19:50:24 127.0.0.1:8083
2022/04/06 19:50:24 127.0.0.1:8082
2022/04/06 19:50:24 127.0.0.1:8083
2022/04/06 19:50:24 127.0.0.1:8082
2022/04/06 19:50:24 127.0.0.1:8083
2022/04/06 19:50:24 127.0.0.1:8082
2022/04/06 19:50:24 127.0.0.1:8083
2022/04/06 19:50:24 127.0.0.1:8082
2022/04/06 19:50:24 127.0.0.1:8083
-------- node 调用次数 --------
2022/04/06 19:50:24 node: 127.0.0.1:8081 useCount: 1
2022/04/06 19:50:24 node: 127.0.0.1:8082 useCount: 5
2022/04/06 19:50:24 node: 127.0.0.1:8083 useCount: 5
第一个 node 宕机,这一段输出了宕机状态
2022/04/06 19:28:48 127.0.0.1:8081
2022/04/06 19:28:48 http://127.0.0.1:8081 is dead.
从这一段可以看出节点服务是被交替调用
2022/04/06 19:28:48 127.0.0.1:8082
2022/04/06 19:28:48 127.0.0.1:8083
2022/04/06 19:28:48 127.0.0.1:8082
2022/04/06 19:28:48 127.0.0.1:8083
2022/04/06 19:28:48 127.0.0.1:8082
2022/04/06 19:28:48 127.0.0.1:8083
2022/04/06 19:28:48 127.0.0.1:8082
2022/04/06 19:28:48 127.0.0.1:8083
2022/04/06 19:28:48 127.0.0.1:8082
2022/04/06 19:28:48 127.0.0.1:8083
在这一段可以看出 node 1 被调用了一次,而后递归调用自己,请求分别被 node2 和 node3 各调用 5 次。
说明我们的轮询调度算法是成功的,大家可以复制代码,自己尝试运行下
2022/04/06 19:28:48 node: 127.0.0.1:8081 useCount: 1
2022/04/06 19:28:48 node: 127.0.0.1:8082 useCount: 5
2022/04/06 19:28:48 node: 127.0.0.1:8083 useCount: 5
随机轮询
随机轮询算法也非常的 easy
我们在 rrlbHandle
函数下面添加如下函数
// 随机轮询算法
func rrrlbHandler(w http.ResponseWriter, r *http.Request) {
maxLen := len(cfg.Nodes)
// Rand Round Robin
mu.Lock()
idx, _ := rand.Int(rand.Reader, big.NewInt(int64(maxLen))) // 获取随机数
currentNode := cfg.Nodes[int(idx.Int64())%maxLen] // 获取随机节点
if currentNode.GetIsDead() {
idx, _ = rand.Int(rand.Reader, big.NewInt(int64(maxLen)))
currentNode = cfg.Nodes[int(idx.Int64())%maxLen]
}
currentNode.useCount++
targetURL, err := url.Parse("http://" + cfg.Nodes[int(idx.Int64())%maxLen].URL)
log.Println(targetURL.Host)
if err != nil {
log.Fatal(err.Error())
}
mu.Unlock()
reverseProxy := httputil.NewSingleHostReverseProxy(targetURL)
reverseProxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, e error) {
// NOTE: It is better to implement retry.
log.Printf("%v is dead.", targetURL)
currentNode.SetDead(true)
rrrlbHandler(w, r)
}
reverseProxy.ServeHTTP(w, r)
}
锁机轮询的核心算法如下
idx, _ := rand.Int(rand.Reader, big.NewInt(int64(maxLen))) // 获取随机数
currentNode := cfg.Nodes[int(idx.Int64())%maxLen] // 获取随机节点
if currentNode.GetIsDead() {
idx, _ = rand.Int(rand.Reader, big.NewInt(int64(maxLen)))
currentNode = cfg.Nodes[int(idx.Int64())%maxLen]
}
编写测试代码来测试下
首先修改proxyServerStart
服务函数
func proxyServerStart() {
var err error
go healthCheck()
s := http.Server{
Addr: cfg.Proxy.Url,
// Handler: http.HandlerFunc(rrlbbHandler), // 关闭轮询调度算法
Handler: http.HandlerFunc(rrrlbHandler), // 开启随机轮询调度算法
}
if err = s.ListenAndServe(); err != nil {
log.Fatal(err.Error())
}
}
测试代码与轮询算法测试代码保持不变
测试结果如下:
-------- node 调度顺序--------
2022/04/06 19:49:51 127.0.0.1:8081
2022/04/06 19:49:51 http://127.0.0.1:8081 is dead.
2022/04/06 19:49:51 127.0.0.1:8082
2022/04/06 19:49:51 127.0.0.1:8081
2022/04/06 19:49:51 http://127.0.0.1:8081 is dead.
2022/04/06 19:49:51 127.0.0.1:8082
2022/04/06 19:49:51 127.0.0.1:8083
2022/04/06 19:49:51 127.0.0.1:8083
2022/04/06 19:49:51 127.0.0.1:8082
2022/04/06 19:49:51 127.0.0.1:8083
2022/04/06 19:49:51 127.0.0.1:8083
2022/04/06 19:49:51 127.0.0.1:8081
2022/04/06 19:49:51 http://127.0.0.1:8081 is dead.
2022/04/06 19:49:51 127.0.0.1:8083
2022/04/06 19:49:51 127.0.0.1:8081
2022/04/06 19:49:51 http://127.0.0.1:8081 is dead.
2022/04/06 19:49:51 127.0.0.1:8082
2022/04/06 19:49:51 127.0.0.1:8082
-------- node 调用次数 --------
2022/04/06 19:49:51 node: 127.0.0.1:8081 useCount: 4
2022/04/06 19:49:51 node: 127.0.0.1:8082 useCount: 5
2022/04/06 19:49:51 node: 127.0.0.1:8083 useCount: 5
从测试结果中可以看出,node 调用顺序是随机的,node 调用次数负载到未宕机的 node2、node3 上总计被调用10次
说明我们的算法也是成功的
加权轮询
加权轮询我们也基于轮询的代码来修改
修改配置文件
{
"proxy": {
"url": "127.0.0.1:8080"
},
"nodes": [
{
"url": "127.0.0.1:8081",
"weight": 2
},
{
"url": "127.0.0.1:8082",
"weight": 3
},
{
"url": "127.0.0.1:8083",
"weight": 5
}
]
}
我们再给 Node
的结构体加两个属性
currentWeight
node临时权重effectiveWeight
node有效权重Weight
node权重
type Node struct {
currentWeight int // node临时权重
effectiveWeight int // node有效权重
Weight int `json:"weight"` // node权重
IsDead bool
useCount int
URL string `json:"url"`
mu sync.RWMutex
}
修改 init
函数如下如下代码
func init() {
data, err := ioutil.ReadFile("./config.json")
if err != nil {
log.Fatal(err.Error())
}
json.Unmarshal(data, &cfg)
for _, node := range cfg.Nodes {
node.currentWeight = node.Weight
}
}
修改 rrlbHandler
函数为如下代码
func rrlbHandler(w http.ResponseWriter, r *http.Request) {
mu.Lock()
currentNode := cfg.Next()
targetURL, err := url.Parse("http://" + currentNode.URL)
if err != nil {
log.Fatal(err.Error())
}
log.Println(targetURL.Host)
mu.Unlock()
reverseProxy := httputil.NewSingleHostReverseProxy(targetURL)
reverseProxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, e error) {
// NOTE: It is better to implement retry.
log.Printf("%v is dead.", targetURL)
currentNode.SetDead(true)
rrlbHandler(w, r)
}
reverseProxy.ServeHTTP(w, r)
}
添加 Next
函数代码如下,此函数即为加权轮询核心算法
注意:在获取最大临时权重 node 的过程中我们要保证最大临时权重node的临时权重要持续递减,而且各个node的临时权重要持续递增,以保证调度的平滑性
func (c *Config) Next() *Node {
totalEffectiveWeight := 0
var maxWeightNode *Node
for _, node := range c.Nodes {
// 1.统计所有node有效权证之和
totalEffectiveWeight += node.effectiveWeight
// 2.变更node临时权重=node临时权重+node有效权重
// node临时权重递增,交替加权,以保证调度平滑性
node.currentWeight += node.effectiveWeight
// 3.node有效权重默认与node权临时重相同,通讯异常时-1,通信成功+1,直到恢复到 weight 大小
if node.effectiveWeight < node.Weight {
if node.GetIsDead() {
node.effectiveWeight--
} else {
node.effectiveWeight++
}
}
// 4.选择最大临时权重node
if maxWeightNode == nil || node.currentWeight > maxWeightNode.currentWeight {
maxWeightNode = node
}
}
if maxWeightNode == nil {
return nil
}
// 5.变更 node临时权重=node临时权重-node有效权重之和
// 最大权重node临时权重递减,交替减权,以保证调度平滑性
maxWeightNode.currentWeight -= totalEffectiveWeight
if maxWeightNode.GetIsDead() {
maxWeightNode = c.Next()
}
maxWeightNode.useCount++
return maxWeightNode
}
测试结果如下:
-------- node 调度顺序--------
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8082
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8082
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8081
2022/04/06 21:50:00 http://127.0.0.1:8081 is dead.
2022/04/06 21:50:00 127.0.0.1:8083
2022/04/06 21:50:00 127.0.0.1:8082
-------- node 调用次数 --------
2022/04/06 21:50:00 node: 127.0.0.1:8081 useCount: 1
2022/04/06 21:50:00 node: 127.0.0.1:8082 useCount: 3
2022/04/06 21:50:00 node: 127.0.0.1:8083 useCount: 7
从结果中可以看出,调度还是比较平滑的,而且对应权重node在调度中调用次数也比较合理