【sentinel】深入浅出之原理篇集群流控之Demo

作者: 一滴水的坚持 | 来源:发表于2019-03-22 17:58 被阅读0次

【sentinel】深入浅出之原理篇集群流控之Demo
【sentinel】深入浅出之原理篇集群流控之原理
深入浅出Redis-redis哨兵集群
sentinel集群流控实践
SpringClouldAlibaba 之 Sentinel流控
Sentinel 1.7.0 发布，支持 Envoy 集群流量控
Sentinel 1.7.0 发布，支持 Envoy 集群流量控
控制台改造与zk接入
【sentinel】深入浅出之原理篇SlotChain
【sentinel】深入浅出之原理篇DegradeSlot

关于集群限流的功能，官方文档写的非常详细:集群流控官方文档
官方文档中有demo，但是隐藏了太多细节，且依赖了nacos，我这里自己写了一个demo。

demo地址: https://github.com/shxz130/sentinel-demo
首先启动一个Server端:

设置FlowRule，并将其设置为Cluster模式
初始化Server(Client和Server都可以共用)
初始化Client，指定Server端地址。
将该节点设置为Server 端，并启动。

 public static void main(String[] args) {
    //初始化限流规则
    initClusterFlowRule();
    //初始化Server
    initServer();
    //初始化Cluster Client规则，指定Server ip和端口
    initClusterClientRule();
    //设置为Server 并启动
    setToServer();
    while (true){
        //等待10秒
        try{
            Thread.sleep(100);
        }catch (Exception e){
            e.printStackTrace();
        }
        try{
            //对资源A限流
            Entry entry=SphU.entry("a");
            System.out.println("pass");
        }catch (Exception e){
            System.out.println("block");
        }
    }
}

在初始化规则的时候，指定集群 or 单机模式，指定限流规则，指定限流类型qps or threads ，最后，加载单机规则，集群NameSpace以及nameSpace对应的集群规则。

private static void initClusterFlowRule(){
   List<FlowRule> flowRules=new ArrayList<FlowRule>();
   FlowRule flowRule=new FlowRule();
   //指定限流规则的Resource
   flowRule.setResource("a");
   //集群限流规则配置
   ClusterFlowConfig clusterFlowConfig=new ClusterFlowConfig();
   //集群失效是否转移
   clusterFlowConfig.setFallbackToLocalWhenFail(true);
   //指定flowId 可以使用IP+进程号区分
   clusterFlowConfig.setFlowId(1L);
   //限流规则 全局总数或者平均分摊
   clusterFlowConfig.setThresholdType(ClusterRuleConstant.FLOW_THRESHOLD_GLOBAL);
   flowRule.setClusterConfig(clusterFlowConfig);
   //是否启用Cluster模式
   flowRule.setClusterMode(true);
   //默认限流规则，具体可以看FlowSlot介绍
   flowRule.setControlBehavior(RuleConstant.CONTROL_BEHAVIOR_DEFAULT);
   //总数
   flowRule.setCount(10);
   //策略
   flowRule.setStrategy(RuleConstant.STRATEGY_DIRECT);
   //限制QPS 也可以指定为线程数
   flowRule.setGrade(RuleConstant.FLOW_GRADE_QPS);
   flowRules.add(flowRule);
   //加载配置
   FlowRuleManager.loadRules(flowRules);
   //注册NameSpace
   ClusterFlowRuleManager.registerPropertyIfAbsent("1-name");
   //NameSpace 下面的规则加载到集群模式
   ClusterFlowRuleManager.loadRules("1-name", flowRules);
}

在初始化Server的时候，指定TokenServer的Ip和端口，并加载nameSpace信息到Cluster。

private static void initServer(){
    //指定提供TokenService的端口和地址
    ServerTransportConfig ServerTransportConfig=new ServerTransportConfig(18730,600);
    //加载配置
    ClusterServerConfigManager.loadGlobalTransportConfig(ServerTransportConfig);
    Set<String> nameSpaceSet=new HashSet<String>();
    nameSpaceSet.add("1-name");
    //服务配置namespace
    ClusterServerConfigManager.loadServerNamespaceSet(nameSpaceSet);
    //ClusterServerConfigManager.loadGlobalFlowConfig();
    //配置了nameSpace对应的ServerFlowConfig
}

初始化客户端的时候，实质是指定客户端规则，配置服务端地址和IP。

private static void initClusterClientRule(){
    //初始化客户端规则
    ClusterClientConfig clusterClientConfig = new ClusterClientConfig();
    //指定获取Token超时时间
    clusterClientConfig.setRequestTimeout(1000);
    //Client指定配置
    ClusterClientConfigManager.applyNewConfig(clusterClientConfig);
    //指定TokenServer的Ip和地址
    ClusterClientAssignConfig clusterClientAssignConfig=new ClusterClientAssignConfig("127.0.0.1",18730);
    //应用
    ClusterClientConfigManager.applyNewAssignConfig(clusterClientAssignConfig);
}

最后，启动服务，则是将上面配置的Server信息对外暴漏服务。

private static void setToServer(){
        ClusterStateManager.setToServer();
}

这里的端口号是设置服务端的端口号,单独启动Server，这里配置1s中10个QPS,模拟1s中10个请求，能看到运行结果如下:

Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:24 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass
Fri Mar 22 17:41:25 CST 2019 pass

结果都是通过的，但并不代表限流生效。将1s中请求10个改为20，也就是超时时间由100改为50，运行结果一半通过，一半阻塞。

Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:10 CST 2019 pass
Fri Mar 22 17:44:10 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 block
Fri Mar 22 17:44:11 CST 2019 pass
Fri Mar 22 17:44:11 CST 2019 block

可以看到，单机限流是生效了。
为了模拟集群环境，启动一个客户端，客户端的配置和服务端区别在于最后启动的不是Server，而是Client。

public static void main(String[] args) {
    initClusterFlowRule();
    initServer();
    initClusterClientRule();
    //设置为客户端
    setToClient();
    while (true){
        try{
            Thread.sleep(100);
        }catch (Exception e){
            e.printStackTrace();
        }
        try{
            Entry entry=SphU.entry("a");
            System.out.println("pass");
        }catch (Exception e){
            System.out.println("block");
        }
    }
}

设置客户端方式和服务端一致。

private static void setToClient(){
    //设置为客户端
    ClusterStateManager.setToClient();
}

先启动服务端。后启动客户端，可以看到服务端日志：

Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:23 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:24 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:25 CST 2019 pass
三月 22, 2019 5:50:25 下午 io.netty.handler.logging.LoggingHandler channelRead
信息: [id: 0xa7c7acc4, L:/0:0:0:0:0:0:0:0:18730] READ: [id: 0x4338a7c7, L:/127.0.0.1:18730 - R:/127.0.0.1:56654]
三月 22, 2019 5:50:25 下午 io.netty.bootstrap.AbstractBootstrap setChannelOption
警告: Unknown channel option 'SO_TIMEOUT' for channel '[id: 0x4338a7c7, L:/127.0.0.1:18730 - R:/127.0.0.1:56654]'
三月 22, 2019 5:50:25 下午 io.netty.handler.logging.LoggingHandler channelReadComplete
信息: [id: 0xa7c7acc4, L:/0:0:0:0:0:0:0:0:18730] READ COMPLETE
Fri Mar 22 17:50:25 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 pass
Fri Mar 22 17:50:30 CST 2019 block
Fri Mar 22 17:50:30 CST 2019 block
Fri Mar 22 17:50:30 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:31 CST 2019 block
Fri Mar 22 17:50:32 CST 2019 block
Fri Mar 22 17:50:32 CST 2019 pass
Fri Mar 22 17:50:32 CST 2019 pass
Fri Mar 22 17:50:32 CST 2019 pass
Fri Mar 22 17:50:32 CST 2019 pass
Fri Mar 22 17:50:32 CST 2019 pass
Fri Mar 22 17:50:32 CST 2019 pass
Fri Mar 22 17:50:32 CST 2019 block
Fri Mar 22 17:50:32 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 block
Fri Mar 22 17:50:33 CST 2019 pass
Fri Mar 22 17:50:34 CST 2019 pass
Fri Mar 22 17:50:34 CST 2019 pass
Fri Mar 22 17:50:34 CST 2019 pass
Fri Mar 22 17:50:34 CST 2019 pass
Fri Mar 22 17:50:34 CST 2019 block

可以看到，当服务端启动的时候，只有服务端一个节点，所以请求全部通过，当客户端联通服务端之后，因为有了客户端连接，所以服务端有block的请求了。
再看一下客户端的日志:

Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 pass
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:26 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 block
Fri Mar 22 17:50:27 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 pass
Fri Mar 22 17:50:28 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 pass
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:29 CST 2019 block
Fri Mar 22 17:50:30 CST 2019 block
Fri Mar 22 17:50:30 CST 2019 block
Fri Mar 22 17:50:30 CST 2019 block
Fri Mar 22 17:50:30 CST 2019 block

可以看到，在同1s的时间里，客户端和服务端的请求通过数加起来就是集群QPS数。

27s 时 客户端pass1 服务端pass9
28s 时 客户端pass1 服务端pass9
29s 时 客户端pass8 服务端pass2

集群限流生效。

【sentinel】深入浅出之原理篇集群流控之Demo
关于集群限流的功能，官方文档写的非常详细:集群流控官方文档官方文档中有demo，但是隐藏了太多细节，且依赖了nac...
【sentinel】深入浅出之原理篇集群流控之原理
在上篇博客中，写了一集群限流的Demo，这篇来分析记录一下集群限流的原理。不管是集群Client，或者是Serve...
深入浅出Redis-redis哨兵集群
深入浅出Redis-redis哨兵集群 1、Sentinel 哨兵 Sentinel（哨兵）是Redis 的高可用...
sentinel集群流控实践
嵌入模式代码示例示例可以参见sentinel-demo-cluster-embedded[https://gi...
SpringClouldAlibaba 之 Sentinel流控
前言上一篇我们将流控规则配置到了nacos 让服务启动时拉取流控规则从而实现持久化但是是有一个缺陷的，毕竟在na...
Sentinel 1.7.0 发布，支持 Envoy 集群流量控
流控降级中间件 Sentinel 1.7.0 版本正式发布，引入了 Envoy 集群流量控制支持、properti...
Sentinel 1.7.0 发布，支持 Envoy 集群流量控
流控降级中间件 Sentinel 1.7.0 版本正式发布，引入了 Envoy 集群流量控制支持、properti...
控制台改造与zk接入
sentinel-dashboard改造注：若要使用集群流控功能，则必须对接动态规则源，否则无法正常使用。 Dy...
【sentinel】深入浅出之原理篇SlotChain
最近在负责做网关类系统，需要考虑做限流熔断功能，基于QPS，基于线程数，对于集群，单机做限流，熔断，而Sentin...
【sentinel】深入浅出之原理篇DegradeSlot
DegradeSlot则通过统计信息，以及预设的规则，来做熔断降级；熔断依赖于响应时长，Qps，失败率。 0 ...