负载均衡算法之二 - 以 Golang 方式

{"type":"doc","content":[{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"过门","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"经过","attrs":{}},{"type":"link","attrs":{"href":"https://hedzr.com/golang/algorithm/go-load-balancer-1","title":"","type":null},"content":[{"type":"text","text":"上一篇","attrs":{}}]},{"type":"text","text":"对基本算法的列举之后，我们注意到基本算法的堆叠是个比较重要的特性。此外，怎么样对 factor 做约束也是一个比较重要的特性，因为它可以帮助决定堆叠后的 LB 如何完成第二级选择。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"建设类库","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以最终，我建立了一个 golang 的 lb 类库：","attrs":{}},{"type":"link","attrs":{"href":"https://github.com/hedzr/lb","title":"","type":null},"content":[{"type":"text","text":"这里","attrs":{}}]},{"type":"text","text":"。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其实 lb 类库真的是多的要命了，不过我也有特定的目的（前面、上一篇也都有提到过）：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"线程安全","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"可堆叠的设计","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"可以对 factor 进行约束","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所谓线程安全无外乎解决 go routines data racing 问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可堆叠则是指基本算法应该能被堆叠起来，形成复合的定制的 LB 算法。比如说将 wrr 和 random 堆叠起来可以形成加权随机算法；或者为一致性hash增加权重也不是不可能。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"对 factor 进行约束是我们的设计目标之一，这是为了让调用者能够传入特定的二级算法的解释器，它将会解释满足了什么样的约束条件之后就可以抽出什么样的结果。它也是为了完整的可堆叠的组成成分之一。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外，算法是要可扩充的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后，由于算法研究是一种非正式的研究行为，所以这次不打算针对高频交易进行优化，因为那样做了之后往往都会面目全非了。尽管如此，由于对每种算法有所选择的原因，所以性能还是很不错的，为免竞争，不做性能测试了。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"可嵌套的设计","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"怎么样设计我们的 LB 库，使得基本算法，特别是加权这样的特性能够嵌套在别的基本算法之上，形成复合的 LB 算法呢？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"例如加权的随机算法，它和加权的轮询算法确确实实还是有区别的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"基本上来说，嵌套套娃的实作已经涉及到强人工智能的领域，也就是我如何创建我自己的终极命题。所以呢，WTF，我只是胡说在八道，其实其关键就在于让 child interface 包含一个 到 parent interface 的嵌入，从而达到隐含的等同性的目的。我这就是传授了 Golang 中的架构设计之终极大法了哦。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"说得有点绕。具体到我们的 LB 设计体系中呢，注意到我们认为你会给 Balancer 添加一堆 Peer(s)，然后在它们上做均衡性计算，计算的依据之一为 Factor 因素，当然，对于轮询和随机这些算法来说并不需要 factor 的参与。所以，当我们想要在 Random Balancer 上叠加 Weighted 算法时，我们需要一个承上启下的 A2B interface，它是 Weighted 算法的 Factor，又是 Random 算法的 Peer，这就解决问题了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"然后在我们的每一个 Next 中会鉴别 next-picked peer 是不是也是一个 BalancerLite，如果是的话就递归进去，从而达到套娃的目的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"比如说，我们的最终 random 算法是这样子的（wrr 中的 Next 也是相似的）：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"func (s *randomS) Next(factor lbapi.Factor) (next lbapi.Peer, c lbapi.Constraintenable) {\n next = s.miniNext()\n if fc, ok := factor.(lbapi.FactorComparable); ok {\n next, c, ok = fc.ConstrainedBy(next)\n } else if nested, ok := next.(lbapi.BalancerLite); ok {\n next, c = nested.Next(factor)\n }\n return\n}\n\nfunc (s *randomS) miniNext() (next lbapi.Peer) {\n s.rw.RLock()\n defer s.rw.RUnlock()\n\n l := int64(len(s.peers))\n ni := atomic.AddInt64(&s.count, inRange(0, l)) % l\n next = s.peers[ni]\n return\n}\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"它一方面做了 BalancerLite 类型诊断然后递归到 child.Next，所以当你在 Balancer.Add(peers) 的时候如果添加对是一个 Balancer 耦合的 Peer 对象的话，嵌套堆叠就能够顺利地级联下去。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另一方面，Next 还针对可被约束的 Factor（即 FactorComparable）做一个约束限定操作（即 ConstrainedBy），这个约束限定操作可以被你用于增强的 LB 目的上。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面就会介绍如何在我们这套 LB 类库上做扩展和定制。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"额外的算法实现","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在这里呢，我们挨个介绍一下两种扩展思路的实例。它们都包含在正式的 LB 库 release 中，一面是做示范，一面是免去你可能的重复编码的无谓浪费——无须担心，实际的生产中各种变态的需求我都有遇到过，你完全不必担心没得代码可写，通用类库怎么也不能替你干完一切事的。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"加权随机算法：对 random 做 weighted 堆叠","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"既然 wrr 能够加权，random 能够随机，那么借助于已经建设好的基础设施我们就可以堆叠两者。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在这个需求里，需要用到 Next 中的这个分支所提供的能力：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":" if nested, ok := next.(lbapi.BalancerLite); ok {\n next, c = nested.Next(factor)\n }\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们的实现方法是，制作一个新的 Peer 实现，让它是一个 random lb，并且能够提供一个权重值，因为首先来讲 wrr 需要我们添加给它的都是 WeightedPeer，这就是代码中的 wpPeer struct，它实现了 WBPeer interface。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"package wrandom\n\nimport (\n \"github.com/hedzr/lb/lbapi\"\n \"github.com/hedzr/lb/wrr\"\n)\n\nfunc New(opts ...lbapi.Opt) lbapi.Balancer { return wrr.New(opts...) }\n\nfunc WithWeightedBalancedPeers(peers ...WBPeer) lbapi.Opt {\n return func(balancer lbapi.Balancer) {\n for _, b := range peers {\n if bp, ok := b.(lbapi.WeightedPeer); ok {\n balancer.Add(bp)\n }\n }\n }\n}\n\nfunc NewPeer(weight int, gen func(opts ...lbapi.Opt) lbapi.Balancer, opts ...lbapi.Opt) WBPeer {\n return &wpPeer{\n weight: weight,\n lb: gen(opts...),\n }\n}\n\n// WBPeer is a weighted, balanced peer.\ntype WBPeer interface {\n lbapi.WeightedPeer\n lbapi.Balancer\n}\n\ntype wpPeer struct {\n weight int\n lb lbapi.Balancer\n}\n\nfunc (w *wpPeer) String() string { return \"wpBeer\" }\n\nfunc (w *wpPeer) Weight() int { return w.weight }\n\nfunc (w *wpPeer) Next(factor lbapi.Factor) (next lbapi.Peer, c lbapi.Constrainable) {\n return w.lb.Next(factor)\n}\nfunc (w *wpPeer) Count() int { return w.lb.Count() }\nfunc (w *wpPeer) Add(peers ...lbapi.Peer) { w.lb.Add(peers...) }\nfunc (w *wpPeer) Remove(peer lbapi.Peer) { w.lb.Remove(peer) }\nfunc (w *wpPeer) Clear() { w.lb.Clear() }\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"按照上一篇中提供的 lbapi 的 interfaces 定义，在示例中我们构造了一个 wbPeer 的 lbapi.Peer 实现。L38 实现了 Peer，L40 帮助实现了 WeightedPeer，而 L42 则完成了 Balancer 的实现。注意到 wpPeer 结构中包含一个 random balancer 的实例 w.lb。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"codeinline","content":[{"type":"text","text":"WithWeightedBalancedPeers(peers ...WBPeer)","attrs":{}}],"attrs":{}},{"type":"text","text":" 需要一组通过 NewPeer 所创建的 Peer 对象，这些对象最后被 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"New","attrs":{}}],"attrs":{}},{"type":"text","text":" 用来构造一个 wrr LB。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以我们可以这样使用它：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"package wrandom_test\n\nimport (\n \"github.com/hedzr/lb/lbapi\"\n \"github.com/hedzr/lb/random\"\n \"github.com/hedzr/lb/wrandom\"\n \"testing\"\n)\n\ntype exP string\n\nfunc (s exP) String() string { return string(s) }\n\nfunc withPeers(peers ...lbapi.Peer) lbapi.Opt {\n return func(balancer lbapi.Balancer) {\n for _, p := range peers {\n balancer.Add(p)\n }\n }\n}\n\nfunc TestWR1(t *testing.T) {\n peer1, peer2, peer3 := exP(\"172.16.0.7:3500\"), exP(\"172.16.0.8:3500\"), exP(\"172.16.0.9:3500\")\n peer4, peer5 := exP(\"172.16.0.2:3500\"), exP(\"172.16.0.3:3500\")\n\n lb := wrandom.New(\n wrandom.WithWeightedBalancedPeers(\n wrandom.NewPeer(3, random.New, withPeers(peer1, peer2, peer3)),\n wrandom.NewPeer(2, random.New, withPeers(peer4, peer5)),\n ),\n )\n\n sum := make(map[lbapi.Peer]int)\n\n for i := 0; i < 5000; i++ {\n p, _ := lb.Next(lbapi.DummyFactor)\n sum[p]++\n }\n\n // results\n for k, v := range sum {\n t.Logf(\"%v: %v\", k, v)\n }\n}\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，这里的代码是为了测试使用的。在业务代码中你可以将上述的一体化初始化语句拆分为多步骤的，在运行时刻对 wpPeer 对象做 Add/Remove 来增减其从属的后端而不是直接加入 peer1, peer2, ..., peer5。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一个参考结果是：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"bash"},"content":[{"type":"text","text":" wr_test.go:42: 172.16.0.8:3500: 1018\n wr_test.go:42: 172.16.0.3:3500: 1012\n wr_test.go:42: 172.16.0.9:3500: 1026\n wr_test.go:42: 172.16.0.2:3500: 988\n wr_test.go:42: 172.16.0.7:3500: 956\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"5 个后端节点最终","attrs":{}},{"type":"text","marks":[{"type":"italic","attrs":{}}],"text":"均分","attrs":{}},{"type":"text","text":"了全部请求，这就是我们想要的结果。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同样的道理，随机算法不必有绝对的均分，允许一定的统计学上的偏差。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"加权的版本比较算法：对 factor 做约束","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"任意组合两种已有的 balancer 算法，上一节已经作出了示范，应该说还算是比较容易实作的吧。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"现在来制作一个全新的加权算法 WV。它包含一个 semver 比较器集合，我们认为调用者将会传入一组带有具体版本号的 host+port 集合作为 factor ，WV 要做的就是：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"按权重比例抽出某一个 semver 比较器 X","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"将 factor 传入 X，从而挑选出一个 host+port+version 使得其 version 满足 X 所约束的条件。","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们使用了 \"github.com/Masterminds/semver\" 来做 semver 相关操作。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"New","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先需要一个 wrr 及其 peers 的专属构建器：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"package version\n\nfunc New(opts ...lbapi.Opt) lbapi.Balancer { return wrr.New(opts...) }\n\nfunc WithConstrainedPeers(cs ...lbapi.Constrainable) lbapi.Opt {\n return func(balancer lbapi.Balancer) {\n for _, vp := range cs {\n balancer.Add(vp)\n }\n }\n}\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"它看起来平平无奇，除了使用一个 Constrainable 接口的 peer 输入之外。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以我们将会为 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"version.New(version.WithConstrainedPeers(...))","attrs":{}}],"attrs":{}},{"type":"text","text":" 传入一堆 Constrainable 这样的 peers。回顾一下 Constrainable 的定义，它本身就是满足 Peer 接口的。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"constrainablePeer","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"现在来实现这样的特殊 peer：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"package version\n\nimport (\n \"fmt\"\n \"github.com/Masterminds/semver\"\n \"github.com/hedzr/lb/lbapi\"\n)\n\nfunc NewConstrainablePeer(constraints string, weight int) (peer lbapi.Constrainable) {\n p := &constrainablePeer{\n constraints: constraints,\n weight: weight,\n constraintsObj: nil,\n }\n vc, err := semver.NewConstraint(constraints)\n if err == nil {\n p.constraintsObj = vc\n peer = p\n }\n return\n}\n\ntype constrainablePeer struct {\n constraints string\n weight int\n constraintsObj *semver.Constraints\n}\n\nfunc (s *constrainablePeer) String() string { return s.constraints }\nfunc (s *constrainablePeer) Weight() int { return s.weight }\nfunc (s *constrainablePeer) Constraints() *semver.Constraints { return s.constraintsObj }\nfunc (s *constrainablePeer) CanConstrain(o interface{}) (yes bool) {\n _, yes = o.(*semver.Version)\n return\n}\nfunc (s *constrainablePeer) Check(factor interface{}) (satisfied bool) {\n if s.constraintsObj == nil {\n var err error\n s.constraintsObj, err = semver.NewConstraint(s.constraints)\n if err != nil {\n fmt.Printf(\"illegal constraints: %q. %v\\n\", s.constraints, err)\n }\n }\n\n if v, ok := factor.(*semver.Version); ok {\n satisfied = s.constraintsObj.Check(v)\n } else if v, ok := factor.(interface{ Version() *semver.Version }); ok {\n satisfied = s.constraintsObj.Check(v.Version())\n }\n return\n}\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"codeinline","content":[{"type":"text","text":"constrainablePeer","attrs":{}}],"attrs":{}},{"type":"text","text":" 的关键之处在于它的 Check 函数，在这里它会试图解开 factor 中包含的 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"*semver.Version","attrs":{}}],"attrs":{}},{"type":"text","text":" 值 V，然后运用约束器 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"s.constraintsObj","attrs":{}}],"attrs":{}},{"type":"text","text":" 来检查 V 是不是满足约束条件。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"翻译成直白的话就是：我是版本号 1.3，我是不是 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"< 1.3.x","attrs":{}}],"attrs":{}},{"type":"text","text":" 啊？","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"constrainablePeer","attrs":{}}],"attrs":{}},{"type":"text","text":" 就会说，是的。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"NewConstrainablePeer","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"constrainablePeer","attrs":{}}],"attrs":{}},{"type":"text","text":" 准备了一个公开函数 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"NewConstrainablePeer","attrs":{}}],"attrs":{}},{"type":"text","text":"，所以：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"var testConstraints = []lbapi.Constrainable{\n version.NewConstrainablePeer(\"<= 1.1.x\", 2),\n version.NewConstrainablePeer(\"^1.2.x\", 4),\n version.NewConstrainablePeer(\"^2.x\", 11),\n version.NewConstrainablePeer(\"^3.x\", 3),\n}\n\n lb := version.New(version.WithConstrainedPeers(testConstraints...))\n","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"TestCase","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"现在需要的是准备好 factor，然后来做测试：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"func initFactors() version.BackendsFactor {\n fa := version.NewBackendsFactor(rr.New)\n fa.AddPeers(\n version.NewBackendFactor(\"1.1\", \"172.16.0.6:3500\"),\n version.NewBackendFactor(\"1.3\", \"172.16.0.7:3500\"),\n version.NewBackendFactor(\"2.0\", \"172.16.0.8:3500\"),\n version.NewBackendFactor(\"3.13\", \"172.16.0.9:3500\"),\n )\n return fa\n}\n\nfactor := initFactors()\npeer, c := lb.Next(factor)\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在这里我们略过了 NewBackendsFactor 以及 NewBackendFactor 的大部分实现细节，其中值得特别提及的是 fs 也就是 backendsFactor 的 ConstrainedBy：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"func (fa *backendsFactor) ConstrainedBy(constraints interface{}) (peer lbapi.Peer, c lbapi.Constrainable, satisfied bool) {\n if cc, ok := constraints.(lbapi.Constrainable); ok {\n var lb lbapi.Balancer\n \n // for this object cc, build a lb and associate with it\n fa.crw.RLock()\n if _, ok := fa.constraints[cc]; !ok {\n fa.crw.RUnlock()\n\n lb = fa.generator()\n fa.crw.Lock()\n fa.constraints[cc] = lb\n fa.crw.Unlock()\n } else {\n lb = fa.constraints[cc]\n fa.crw.RUnlock()\n lb.Clear()\n }\n\n // find all satisfied backends/peers and add them into lb\n for _, f := range fa.backends {\n if cc.Check(f) {\n satisfied = true\n lb.Add(f)\n }\n }\n\n // now, pick up the next peer of them\n peer, c = lb.Next(lbapi.DummyFactor)\n if c == nil {\n c = cc\n }\n }\n return\n}\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一般地讲，在这里 fa.generator 应该是一个 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"rr.New","attrs":{}}],"attrs":{}},{"type":"text","text":"（通过 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"fa := version.NewBackendsFactor(rr.New)","attrs":{}}],"attrs":{}},{"type":"text","text":"），也即 lb 会被创建为一个轮询算法的 LB，或者别的亦可。然后全部满足约束条件（通过 Check）的后端将被当作是一个 peers 集合让 lb 进行选择。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"感兴趣的朋友可以去到源码查阅。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"出于代码逻辑清晰的缘故，上面采用了代价较高的运算，也就是每次都对全部 fa.backends 做一次 Check 比较。可想而知对于大规模的后端它不是好的方式。以后将会在此增加 cache 并做增量式添加删除，以便尽可能地消除 Check all。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":"小小结","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有何用意？这个加权的版本范围比较算法，其实是我以前做 api gw 时的一个念头，其目的在于做到整个自研框架的灰度上线测试。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，作为一个完整的框架要达成灰度上线目标，需要的是全方位的适配。至少会包含这些：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"可以按照版本范围的约束表达式来进行权重分配","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"能够透明地完成 incoming requests 的 dispatch","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","text":"利用 cmdr 锁提供的配置文件自动监控、自动载入和合并、自动触发变更机制，以便在 microservice 在线的状态下实时地调整权重分配设定","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"如果没有配置文件合并能力，则需要考虑采用诸如 redis 缓存的配置项等手段来保证实时权重分配。","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"需要能够综合多种负载均衡算法，特别是正确地抽取请求上下文中的 Properties 来完成分发决策","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，真正全功能的 API GW，已知的能开箱即用的大概还没有，能够通过代码调整来适配的可能都有限，或者是需要在系统的各种角落做不知名的 hack 才行。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"最后","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最后，把所有的 LB 算法收归到上一级 package 中，我们用一个 map，这个 map 配备了一颗 RWMutex 来防止你会在多线程的环境中使用它。但是实际上我们认为你应该事先建立一个 lb 的实例，之后动态地 add/remove 它的 peers 就好了，此外，我们也认为你应该是在 app 一开始的时候就已经注册了自行定制的 balancer 算法及其 generator 的——也就是说，其实这颗锁意思有限。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"好，在上一级的总领中，代码片段是这样：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"func New(algorithm string, opts ...lbapi.Opt) lbapi.Balancer {\n kbs.RLock()\n defer kbs.RUnlock()\n if g, ok := knownBalancers[algorithm]; ok {\n return g(opts...)\n }\n log.Fatalf(\"unknown/unregistered balancer and generator: %q\", algorithm)\n return nil // unreachable\n}\n\n// WithPeers adds the initial peers.\nfunc WithPeers(peers ...lbapi.Peer) lbapi.Opt {\n return func(balancer lbapi.Balancer) {\n for _, p := range peers {\n balancer.Add(p)\n }\n }\n}\n\n// Register assign a (algorithm, generator) pair.\nfunc Register(algorithm string, generator func(opts ...lbapi.Opt) lbapi.Balancer) {\n kbs.Lock()\n defer kbs.Unlock()\n knownBalancers[algorithm] = generator\n}\n\n// Unregister revoke a (algorithm, generator) pair.\nfunc Unregister(algorithm string) {\n kbs.Lock()\n defer kbs.Unlock()\n delete(knownBalancers, algorithm)\n}\n\nconst (\n // Random algorithm\n Random = \"random\"\n // RoundRobin algorithm\n RoundRobin = \"round-robin\"\n // WeightedRoundRobin algorithm\n WeightedRoundRobin = \"weighted-round-robin\"\n // ConsistentHash algorithm\n ConsistentHash = \"consistent-hash\"\n // WeightedRandom algorithm\n WeightedRandom = \"weighted-random\"\n // VersioningWRR algorithm\n VersioningWRR = \"versioning-wrr\"\n)\n\nfunc init() {\n kbs.Lock()\n defer kbs.Unlock()\n\n knownBalancers = make(map[string]func(opts ...lbapi.Opt) lbapi.Balancer)\n\n knownBalancers[Random] = random.New\n knownBalancers[RoundRobin] = rr.New\n knownBalancers[WeightedRoundRobin] = wrr.New\n knownBalancers[ConsistentHash] = hash.New\n\n knownBalancers[WeightedRandom] = wrandom.New\n\n knownBalancers[VersioningWRR] = version.New\n}\n\nvar knownBalancers map[string]func(opts ...lbapi.Opt) lbapi.Balancer\nvar kbs sync.RWMutex\n","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"用法","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个类库允许你 register 自己的 lb 算法进去，目的在于让调用者能够有更统一的界面。至于使用某个 lb 算法是很简单的：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"go"},"content":[{"type":"text","text":"package main\n\nimport (\n \"fmt\"\n lb \"github.com/hedzr/lb\"\n \"github.com/hedzr/lb/lbapi\"\n)\n\nfunc main() {\n b := lb.New(lb.RoundRobin)\n\n b.Add(exP(\"172.16.0.7:3500\"), exP(\"172.16.0.8:3500\"), exP(\"172.16.0.9:3500\"))\n sum := make(map[lbapi.Peer]int)\n for i := 0; i < 300; i++ {\n p, _ := b.Next(lbapi.DummyFactor)\n sum[p]++\n }\n \n for k, v := range sum {\n fmt.Printf(\"%v: %v\\n\", k, v)\n }\n}\n\ntype exP string\n\nfunc (s exP) String() string { return string(s) }\n","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"根据你的实际场景，你需要自己的 exP 实现（它需要实现 ","attrs":{}},{"type":"codeinline","content":[{"type":"text","text":"lbapi.Peer","attrs":{}}],"attrs":{}},{"type":"text","text":"），但为了便利于你的使用，一个 Peer 只需要实现了 String() string 接口就够了。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"结束","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，在这次放出的 LB 类库中，我们提供了一堆基本算法以及复合算法：","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"numberedlist","attrs":{"start":1,"normalizeStart":1},"content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":1,"align":null,"origin":null},"content":[{"type":"text","text":"随机","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":2,"align":null,"origin":null},"content":[{"type":"text","text":"轮询","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":3,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"del","attrs":{}}],"text":"最少连接数","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":4,"align":null,"origin":null},"content":[{"type":"text","text":"hashing","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":5,"align":null,"origin":null},"content":[{"type":"text","text":"加权轮询","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":6,"align":null,"origin":null},"content":[{"type":"text","text":"加权随机","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":7,"align":null,"origin":null},"content":[{"type":"text","text":"加权的版本范围比较","attrs":{}}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而且，我们认为已经提供了充足的基础设施以便利于你在一个 unified 的框架下面拓展负载均衡算法达成你的实际需要。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"话说昨天刷 github，发现 profile page 大变样了，有 explore，trending 之类的 pages。今天现在去看又恢复老样子了。我难不成是幻觉了，还是说是昨晚想着这第二篇该怎么写就睡着了在 dreaming？","attrs":{}}]}],"attrs":{}}]}