Add Maglev hashing LB algorithm

[shawnh2]

• add Maglev hash load balancing algorithm
• rename original consistent hash to ring-hash

[codecov-commenter]

Codecov Report

Merging #554 (b80db56) into develop (52728f8) will increase coverage by 0.04%.
The diff coverage is 76.43%.

❗ Current head b80db56 differs from pull request most recent head f0022c6. Consider uploading reports for the commit f0022c6 to get more accurate results

@@             Coverage Diff             @@
##           develop     #554      +/-   ##
===========================================
+ Coverage    54.71%   54.75%   +0.04%     
===========================================
  Files          667      669       +2     
  Lines        78143    78323     +180     
===========================================
+ Hits         42756    42886     +130     
- Misses       31733    31778      +45     
- Partials      3654     3659       +5     

Impacted Files	Coverage Δ
pixiu/pkg/server/cluster_manager.go	36.84% <0.00%> (ø)
...xiu/pkg/cluster/loadbalancer/maglev/maglev_hash.go	62.96% <62.96%> (ø)
...xiu/pkg/cluster/loadbalancer/maglev/permutation.go	80.71% <80.71%> (ø)
...xiu/pkg/cluster/loadbalancer/ringhash/ring_hash.go	71.42% <84.21%> (ø)

... and 19 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[baerwang]

分配好像不均衡，result map[1:6 2:2 3:4 4:5 5:3]

func TestMaglevHash(t *testing.T) {

	nodeCount := 5

	nodes := make([]*model.Endpoint, 0, nodeCount)

	for i := 1; i <= nodeCount; i++ {
		name := strconv.Itoa(i)
		nodes = append(nodes, &model.Endpoint{ID: name, Name: name,
			Address: model.SocketAddress{Address: "192.168.1." + name, Port: 1000 + i}})
	}

	cluster := &model.ClusterConfig{
		Name:           "test-cluster",
		Endpoints:      nodes,
		LbStr:          model.LoadBalanceMaglevHashing,
		ConsistentHash: model.ConsistentHash{ReplicaFactor: 10},
	}
	cluster.CreateConsistentHash()

	var (
		hashing = MaglevHash{}
		path    string
	)

	record := make(map[string]int, nodeCount)

	for i := 1; i <= 20; i++ {
		path = fmt.Sprintf("/pixiu?total=%d", i)
		handler := hashing.Handler(cluster, &http.HttpContext{Request: &stdHttp.Request{Method: stdHttp.MethodGet, RequestURI: path}})
		if _, ok := record[handler.ID]; ok {
			record[handler.ID]++
			continue
		}
		record[handler.ID] = 1
	}

	t.Log(record)
}

[shawnh2]

分配好像不均衡，result map[1:6 2:2 3:4 4:5 5:3]

maglev hash 算法要求的是 LookUpTable 的表长度（也就是其结构体内 size 字段的值）必须是个质数，这样可以保证均衡。

但在目前的实现中，这一点我没有去保证，我来补充一下。

[shawnh2]

分配好像不均衡，result map[1:6 2:2 3:4 4:5 5:3]

maglev hash 算法要求的是 LookUpTable 的表长度（也就是其结构体内 size 字段的值）必须是个质数，这样可以保证均衡。

但在目前的实现中，这一点我没有去保证，我来补充一下。

fixed, @baerwang PTAL

[mark4z]

Using a simple benchmark testing tool https://github.com/mark4z/hey to test the effect of load balancing, 1 of 5 endpoints always can not receive requests, and starts to get 503 frequently after a period of time.

[mark4z]

---
static_resources:
  listeners:
    - name: "net/http"
      protocol_type: "HTTP"
      address:
        socket_address:
          address: "0.0.0.0"
          port: 8888
      filter_chains:
        filters:
          - name: dgp.filter.httpconnectionmanager
            config:
              route_config:
                routes:
                  - match:
                      prefix: "/count"
                    route:
                      cluster: "count"
                      cluster_not_found_response_code: 505
              http_filters:
                - name: dgp.filter.http.httpproxy
                  config:
      config:
        idle_timeout: 5s
        read_timeout: 5s
        write_timeout: 5s
  clusters:
    - name: "count"
      lb_policy: "MaglevHashing"
      endpoints:
        - id: 1
          socket_address:
            address: localhost
            port: 8080
        - id: 2
          socket_address:
            address: localhost
            port: 8081
        - id: 3
          socket_address:
            address: localhost
            port: 8082
        - id: 4
          socket_address:
            address: localhost
            port: 8083
        - id: 5
          socket_address:
            address: localhost
            port: 8084
  shutdown_config:
    timeout: "60s"
    step_timeout: "10s"
    reject_policy: "immediacy"

[shawnh2]

Using a simple benchmark testing tool https://github.com/mark4z/hey to test the effect of load balancing, 1 of 5 endpoints always can not receive requests, and starts to get 503 frequently after a period of time.

此次测试，由于该版代码并没有在 pluginregistry 中注册，所以压测的实质上是默认的 rand 策略。至于为什么总有一个节点接收不到请求，我看了一下是 rand 的随机区间设置的有问题，目前已更正。

[shawnh2]

new benchmark result on Maglev hash, using same config as #554 (comment)

endpoints 1 recv: 5847
endpoints 2 recv: 5840
endpoints 3 recv: 5770
endpoints 4 recv: 5782
endpoints 5 recv: 5675

[sonarcloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
7 Code Smells

No Coverage information
1.9% Duplication

[mark4z]

new benchmark result on Maglev hash, using same config as #554 (comment)

endpoints 1 recv: 5847 endpoints 2 recv: 5840 endpoints 3 recv: 5770 endpoints 4 recv: 5782 endpoints 5 recv: 5675

nice job