Golang实践录:使用gin框架实现转发功能:一些负载均衡

这是我参与11月更文挑战的第17天,活动详情查看:2021最后一次更文挑战

近段时间需要实现一个转发 post 请求到指定后端服务的小工具,由于一直想学习 gin 框架,所以就使用这个框架进行尝试,预计会产生几篇文章。本文研究一些负载均衡算法的实现。

概述

本文实现的负载均衡纯粹是为了笔者的一个念想,并不具有实际指导意义。

本文假定有一个后端服务 URL 的数组,因此,在实现上,仅是输出数组的索引值。权重数组每个索引值对应一个后端服务。比如”1 3 3”,表示有3台服务器,第1台权重为1,第2、3台权重均为3。

算法实现及测试结果

负载均衡有很多类型,如:随机、加权随机;简单轮询、加权轮询、平滑加权轮询,等。本文仅实现几种轮询算法,并且按请求次序递增,不使用随机数。

简单轮询算法

算法描述:

按请求先后轮询服务器。接收到第一次请求,转发到第1台服务器,第二次请求,转发到第2台服务器,依次类推,如果轮询到最后一台服务器,再转发第一台。

代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
go复制代码​
var lb_index int = 0

// 简单轮询
func getOneServerUrlForLB_RR(exTime string) (url string) {

  url = ""
  count := conf.TotalPort //len(conf.BackInfo)

  if lb_index >= count {
      lb_index = 0
  }
  fmt.Println("indx: ", lb_index, count)
  url = fmt.Sprintf("http://127.0.0.1:%d", conf.BackPorts[lb_index])
  fmt.Println("got url: ", url)

  lb_index += 1

  return
}

加权轮询算法

一个权重的示意:

3台机器(分别为A、B、C),权重分别为5,3,2。排列(注:按设置的权重先后列出):

1
2
3
4
css复制代码   5       3    2
1       5     8   10
| ----- | --- | -- |
  A       B     C

某一区间大,表示该服务器的权重大。 设一数,其值在1~10之间,即范围在权重总和之内,依次与5、3、2对比,如小则在该区间,如大,则减去前一数,再比对。看落到哪个区域。 举例: 设该数为3,与5对比,小,则落到第1区间,即选择服务器 A。 设该数为7,与5对比,大,取下一区间,即减去5,得2,与3对比,小,则落到第2区间,即选择服务器 B。 设该数为9,与5对比,大,减去5得4,与3对比,大,减去3得1,与2对比,小,第3区间,即选择服务器 C。

注意,我们关注的是某个区间出现的次数,并不关注是哪一个索引。以轮询 10 次为例,只要保证A、B、C服务器分别访问了5、3、2次即可。最简单的算法,就是将这次 10 次依次分配到A、B、C服务器。

代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
go复制代码/*
测试用,假定有3组IP,每组分配好权重,根据请求顺序,按权重分配
可以认为,IP总数为权重总值,只是部分重复了,所以请求循环的次数就是权重总值
*/
// 因为测试,所以直接赋值,测试的总数不超过 conf.TotalPort
// conf.Weights 示例:[3]int{2, 5, 3}, 数组索引0表示第1个后端,1表示第2个后端

func getIndex_WRR(offset int, totalWeight int) (index int) {

   // 如果不指定,使用随机
   if offset == -1 {
       rand.Seed(time.Now().UnixNano())
       offset = rand.Int() % totalWeight
  }
   // fmt.Println("total: ", totalWeight, offset)

   // 注:这里关注的是某个索引值的次数,与顺序无关
   for index, w := range conf.Weights {
       if offset < w {
           return index
      }
       offset -= w
  }

   return
}

func getOneServerUrlForLB_WRR(exTime string) (url string) {

   url = ""
   count := conf.TotalPort

   totalWeight := 0
   for _, w := range conf.Weights {
       totalWeight += w
  }

   count = totalWeight // 注:按权重总数轮询
   if lb_index >= count {
       lb_index = 0
  }
   new_lb_index := getIndex_WRR(lb_index, totalWeight)

   if lb_index == 100 {
       fmt.Println("indx: ", lb_index, new_lb_index, count)
  }
   fmt.Printf("%d ", new_lb_index)

   url = fmt.Sprintf("http://127.0.0.1:%d", conf.BackPorts[new_lb_index])
   fmt.Println("got url: ", url)

   lb_index += 1

   return
}

实验结果:

1
2
3
4
5
6
7
8
bash复制代码./httpforward.exe -w "5 3 2"
0 0 0 0 0 1 1 1 2 2 0 0 0 0 0 1 1 1 2 2

./httpforward.exe -w "1 3 3"
0 1 1 1 2 2 2 0 1 1 1 2 2 2

./httpforward.exe -w "5 3 1"
0 0 0 0 0 1 1 1 2 0 0 0 0 0 1 1 1 2

为了数据可靠,轮询了2遍,下同。

平滑加权轮询算法

上述算法并不考虑服务器处理在效率,比如前面5次均在A服务器,其它服务器均为空闲状态,由此引出平滑加权轮询算法。笔者暂未参透该算法的证明过程,因此本文不涉及。

设置如下变量: 总权重值:totalWeight,其值固定。 固定权重数组:Weights,其值固定。 当前权重数组 CurrentWeights,其值可变。 当前权重数组最大值 maxWeight,CurrentWeights数组的最大值。

算法步骤: 0、某次请求到来。 1、判断当前权重数组(对应下表当前权重1)值是否全为0,如是,则用固定权重初始化之。 2、查获当前权重数组,获取最大值,对应的索引,即为需要返回的服务器。 3、将第2步的索引对应的值,减去权重总和,其它索引的值保持不变。(对应下表当前权重2) 4、 将第3步得到的当前权重数组的每个值,加上对应的固定权重值。 5、回到第1步,重复。

假定有3台服务器A、B、C,其权重依次为 5、1、1,权重总和为7。演算过程如下:

初始时,当前权重为[0,0,0],则用固定权重[5,1,1]初始化之。 此时,最大值为5,索引为0,返回服务器A。 将[5,1,1]的索引0值减去权重总和7,得到[-2,1,1]。 将[-2,1,1]加上[5,1,1],得到新的当前权重[3,2,2]。 此时,最大值为3,索引为0,返回服务器A。 将[3,2,2]的索引0值减去权重总和7,得到[-4,2,2]。 将[-4,2,2]加上[5,1,1],得到新的当前权重[1,3,3]。 (下略)

总体过程如下表:

请求次数 当前权重1 返回服务器 当前权重2
0 [0, 0,0] -- 初始化为[5, 1, 1]
1 [5, 1, 1] A [-2, 1, 1]
2 [3, 2, 2] A [-4, 2, 2]
3 [1, 3, 3] B [1, -4, 3]
4 [6, -3, 4] A [-1, -3, 4]
5 [4, -2, 5] C [4, -2, -2]
6 [9, -1, -1] A [2, -1, -1]
7 [7, 0, 0] A [0, 0, 0]
8 [5, 1, 1] A [-2, 1, 1]

代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
go复制代码// 平滑加权
func getIndex_SWRR(offset int, totalWeight int) (index int) {

   initflag := 0

   // 0.判断是否全为0
   for _, w := range conf.CurrentWeights {
       if w == 0 {
           initflag++
      }
  }
   // 全为0,则需要初始化
   if initflag == len(conf.CurrentWeights) {
       for idx, w := range conf.Weights {
           conf.CurrentWeights[idx] = w
      }
  }

   // 1. 查询当前权重表最大者,并取该索引,即为所需结果
   maxWeight := 0
   for idx, w := range conf.CurrentWeights {
       if w > maxWeight {
           maxWeight = w
           index = idx
      }
  }
   // 2. 最大的那个权重,其值减去总权重
   conf.CurrentWeights[index] = maxWeight - totalWeight
   // fmt.Println("current222: ", conf.CurrentWeights)

   // 3. 新的当前权重,所对应的值加上初始权重
   // 为下次循环计算打下基础
   for idx, w := range conf.Weights {
       conf.CurrentWeights[idx] += w
  }
   return
}

func getOneServerUrlForLB_SWRR(exTime string) (url string) {
   url = ""

   totalWeight := 0

   for _, w := range conf.Weights {
       totalWeight += w
  }

   // 注:按权重总数轮询
   if lb_index >= totalWeight {
       lb_index = 0
  }
   new_lb_index := getIndex_SWRR(lb_index, totalWeight)

   if lb_index == 100 {
       fmt.Println("indx1: ", lb_index, new_lb_index, totalWeight)
  }
   fmt.Printf("%d ", new_lb_index)
   url = fmt.Sprintf("http://127.0.0.1:%d", conf.BackPorts[new_lb_index])
   //fmt.Println("got url: ", url)

   lb_index += 1

   return
}

实验结果:

1
2
3
4
5
6
7
8
9
10
11
bash复制代码./httpforward.exe -w "5 1 1"
0 0 1 0 2 0 0 0 0 1 0 2 0 0

./httpforward.exe -w "1 3 3"
1 2 0 1 2 1 2 1 2 0 1 2 1 2

./httpforward.exe -w "5 3 1"
0 1 0 2 0 1 0 1 0 0 1 0 2 0 1 0 1 0

./httpforward.exe -w "2 5 3"
1 2 0 1 1 2 1 0 2 1 1 2 0 1 1 2 1 0 2 1

从结果上看,权重保持着比例,但响应的服务器分布较平衡,即不会出现权重大的服务器集中处理请求的情况。某次测试 nginx 的负载均衡,似乎默认便是此种方法,但算法实现不同,测试发现,在保持权重比例情况下,不同的服务器出现的顺序不同,未深究原因。

小结

本文主要根据网络的相关资料整理并用 golang 代码实现负载均衡部分算法。

参考

www.cnblogs.com/wsw-seu/p/1…

juejin.cn/post/684490…

本文转载自: 掘金

开发者博客 – 和开发相关的 这里全都有

0%