昆明市着力构筑中华民族共有精力家园

洞悉九：昆明不要过火依靠经历，挣钱往往需求换个脑子前面的八点洞悉谈了许多关于拉美的现实，有好也有坏。

//以分钟级的核算特点为例，力构力看一下时刻窗口初始化进程privatetransientMetricrollingCounterInMinute=newArrayMetric(60,60*1000,false);publicLeapArray(intsampleCount,intintervalInMs){AssertUtil.isTrue(sampleCount>0,bucketcountisinvalid:+sampleCount);AssertUtil.isTrue(intervalInMs>0,totaltimeintervaloftheslidingwindowshouldbepositive);AssertUtil.isTrue(intervalInMs%sampleCount==0,timespanneedstobeevenlydivided);//windowLengthInMs=60*1000/60=1000滑动窗口时刻长度，力构力可见sentinel默许将单位时刻分为了60个滑动窗口进行数据核算this.windowLengthInMs=intervalInMs/sampleCount;//60*1000this.intervalInMs=intervalInMs;//60this.intervalInSecond=intervalInMs/1000.0;//60this.sampleCount=sampleCount;//数组长度60this.array=newAtomicReferenceArray(sampleCount);}/***Getbucketitematprovidedtimestamp.**paramtimeMillisavalidtimestampinmilliseconds*returncurrentbucketitematprovidedtimestampifthetimeisvalid;nulliftimeisinvalid*/publicWindowWrapcurrentWindow(longtimeMillis){if(timeMillisold.windowStart()){/**(old)*B0B1B2NULLB4*|_______||_______|_______|_______|_______|_______||___*...120014001600180020002200timestamp*^*time=1676*startTimeofBucket2:400,deprecated,shouldbereset**Ifthestarttimestampofoldbucketisbehindprovidedtime,thatmeans*thebucketisdeprecated.Wehavetoresetthebuckettocurrent{codewindowStart}.*Notethattheresetandclean-upoperationsarehardtobeatomic,*soweneedaupdatelocktoguaranteethecorrectnessofbucketupdate.**Theupdatelockisconditional(tinyscope)andwilltakeeffectonlywhen*bucketisdeprecated,soinmostcasesitwontleadtoperformanceloss.*/if(updateLock.tryLock()){try{//Successfullygettheupdatelock,nowweresetthebucket.returnresetWindowTo(old,windowStart);}finally{updateLock.unlock();}}else{//Contentionfailed,thethreadwillyielditstimeslicetowaitforbucketavailable.Thread.yield();}}elseif(windowStart。2.2初始化2.2.1Context初始化在初始化slot职责链部分前，筑中族共还履行了context的初始化，筑中族共里边触及几个重要概念，需求解释一下：能够发现在Context初始化的进程中，会把EntranceNode加入到Root子节点中（实践Root自身是一个特别的EntranceNode），并把EntranceNode放到contextNameNodeMap中。

•广泛的开源生态：华民Sentinel供给开箱即用的与其它开源结构/库的整合模块，例如与SpringCloud、ApacheDubbo、gRPC、Quarkus的整合。项目结构以下首要剖析core包里的内容2.1注解进口2.1.1Entry、有精园Context、有精园NodeSphU门面类的办法出参都是Entry，Entry能够理解为每次进入资源的一个凭据，假如调用SphO.entry()或许SphU.entry()能获取Entry方针，代表获取了凭据，没有被限流，不然抛出一个BlockException。•Entry是一个抽象类，昆明CtEntry是Entry的完结，昆明CtEntry持有Context和调用链的信息Context的源码注释如下，ThisclassholdsmetadataofcurrentinvocationNode的源码注释Holdsreal-timestatisticsforresourcesNode中保存了对资源的实时数据的核算，Sentinel中的限流或许降级等功用便是经过Node中的数据进行判别的。

•FlowSlot则用于依据预设的限流规矩以及前面slot核算的状况，力构力来进行流量操控。DefaultControllerOverridepublicbooleancanPass(Nodenode,intacquireCount,booleanprioritized){intcurCount=avgUsedTokens(node);if(curCount+acquireCount>count){if(prioritized&&grade==RuleConstant.FLOW_GRADE_QPS){longcurrentTime;longwaitInMs;currentTime=TimeUtil.currentTimeMillis();waitInMs=node.tryOccupyNext(currentTime,acquireCount,count);if(waitInMs0,statDurationMsshouldbepositive);AssertUtil.assertTrue(maxCountPerStat>=0,maxCountPerStatshouldbe>=0);AssertUtil.assertTrue(queueingTimeoutMs>=0,queueingTimeoutMsshouldbe>=0);this.maxQueueingTimeMs=queueingTimeoutMs;this.count=maxCountPerStat;this.statDurationMs=statDurationMs;//UsenanoSecondswhendurationMs%count!=0orcount/durationMs>1(tobeaccurate)//可见装备限流值count大于1000时useNanoSeconds会是true不然是falseif(maxCountPerStat>0){this.useNanoSeconds=statDurationMs%Math.round(maxCountPerStat)!=0||maxCountPerStat/statDurationMs>1;}else{this.useNanoSeconds=false;}}OverridepublicbooleancanPass(Nodenode,intacquireCount){returncanPass(node,acquireCount,false);}privatebooleancheckPassUsingNanoSeconds(intacquireCount,doublemaxCountPerStat){finallongmaxQueueingTimeNs=maxQueueingTimeMs*MS_TO_NS_OFFSET;longcurrentTime=System.nanoTime();//Calculatetheintervalbetweeneverytworequests.finallongcostTimeNs=Math.round(1.0d*MS_TO_NS_OFFSET*statDurationMs*acquireCount/maxCountPerStat);//Expectedpasstimeofthisrequest.longexpectedTime=costTimeNs+latestPassedTime.get();if(expectedTimethreshold){transformToOpen(curCount);}}staticclassSimpleErrorCounter{privateLongAddererrorCount;privateLongAddertotalCount;publicSimpleErrorCounter(){this.errorCount=newLongAdder();this.totalCount=newLongAdder();}publicLongAddergetErrorCount(){returnerrorCount;}publicLongAddergetTotalCount(){returntotalCount;}publicSimpleErrorCounterreset(){errorCount.reset();totalCount.reset();returnthis;}OverridepublicStringtoString(){returnSimpleErrorCounter{+errorCount=+errorCount+,totalCount=+totalCount+};}}staticclassSimpleErrorCounterLeapArrayextendsLeapArray{publicSimpleErrorCounterLeapArray(intsampleCount,intintervalInMs){super(sampleCount,intervalInMs);}OverridepublicSimpleErrorCounternewEmptyBucket(longtimeMillis){returnnewSimpleErrorCounter();}OverrideprotectedWindowWrapresetWindowTo(WindowWrapw,longstartTime){//Updatethestarttimeandresetvalue.w.resetTo(startTime);w.value().reset();returnw;}}}2.6SystemSlot校验逻辑首要会集在com.alibaba.csp.sentinel.slots.system.SystemRuleManager#checkSystem，筑中族共以下是片段，筑中族共能够看到，作为负载维护规矩校验，完结了集群的QPS、线程、RT（呼应时刻）、体系负载的操控，除体系负载以外，其他核算都是依靠StatisticSlot完结，体系负载是经过SystemRuleManager守时调度SystemStatusListener，经过OperatingSystemMXBean去获取/***Apply{linkSystemRule}totheresource.Onlyinboundtrafficwillbechecked.**paramresourceWrappertheresource.*throwsBlockExceptionwhenanysystemrulesthresholdisexceeded.*/publicstaticvoidcheckSystem(ResourceWrapperresourceWrapper,intcount)throwsBlockException{if(resourceWrapper==null){return;}//Ensurethecheckingswitchison.if(!checkSystemStatus.get()){return;}//forinboundtrafficonlyif(resourceWrapper.getEntryType()!=EntryType.IN){return;}//totalqps此处是拿到某个资源在集群中的QPS总和，相关概念能够会看初始化关于Node的介绍doublecurrentQps=Constants.ENTRY_NODE.passQps();if(currentQps+count>qps){thrownewSystemBlockException(resourceWrapper.getName(),qps);}//totalthreadintcurrentThread=Constants.ENTRY_NODE.curThreadNum();if(currentThread>maxThread){thrownewSystemBlockException(resourceWrapper.getName(),thread);}doublert=Constants.ENTRY_NODE.avgRt();if(rt>maxRt){thrownewSystemBlockException(resourceWrapper.getName(),rt);}//load.BBRalgorithm.if(highestSystemLoadIsSet&&getCurrentSystemAvgLoad()>highestSystemLoad){if(!checkBbr(currentThread)){thrownewSystemBlockException(resourceWrapper.getName(),load);}}//cpuusageif(highestCpuUsageIsSet&&getCurrentCpuUsage()>highestCpuUsage){thrownewSystemBlockException(resourceWrapper.getName(),cpu);}}privatestaticbooleancheckBbr(intcurrentThread){if(currentThread>1&¤tThread>Constants.ENTRY_NODE.maxSuccessQps()*Constants.ENTRY_NODE.minRt()/1000){returnfalse;}returntrue;}publicstaticdoublegetCurrentSystemAvgLoad(){returnstatusListener.getSystemAverageLoad();}publicstaticdoublegetCurrentCpuUsage(){returnstatusListener.getCpuUsage();}publicclassSystemStatusListenerimplementsRunnable{volatiledoublecurrentLoad=-1;volatiledoublecurrentCpuUsage=-1;volatileStringreason=StringUtil.EMPTY;volatilelongprocessCpuTime=0;volatilelongprocessUpTime=0;publicdoublegetSystemAverageLoad(){returncurrentLoad;}publicdoublegetCpuUsage(){returncurrentCpuUsage;}Overridepublicvoidrun(){try{OperatingSystemMXBeanosBean=ManagementFactory.getPlatformMXBean(OperatingSystemMXBean.class);currentLoad=osBean.getSystemLoadAverage();/**JavaDoccopiedfrom{linkOperatingSystemMXBean#getSystemCpuLoad()}:*Returnstherecentcpuusageforthewholesystem.Thisvalueisadoubleinthe[0.0,1.0]interval.*Avalueof0.0meansthatallCPUswereidleduringtherecentperiodoftimeobserved,whileavalue*of1.0meansthatallCPUswereactivelyrunning100%ofthetimeduringtherecentperiodbeing*observed.Allvaluesbetween0.0and1.0arepossibledependingoftheactivitiesgoingoninthe*system.Ifthesystemrecentcpuusageisnotavailable,themethodreturnsanegativevalue.*/doublesystemCpuUsage=osBean.getSystemCpuLoad();//calculateprocesscpuusagetosupportapplicationrunningincontainerenvironmentRuntimeMXBeanruntimeBean=ManagementFactory.getPlatformMXBean(RuntimeMXBean.class);longnewProcessCpuTime=osBean.getProcessCpuTime();longnewProcessUpTime=runtimeBean.getUptime();intcpuCores=osBean.getAvailableProcessors();longprocessCpuTimeDiffInMs=TimeUnit.NANOSECONDS.toMillis(newProcessCpuTime-processCpuTime);longprocessUpTimeDiffInMs=newProcessUpTime-processUpTime;doubleprocessCpuUsage=(double)processCpuTimeDiffInMs/processUpTimeDiffInMs/cpuCores;processCpuTime=newProcessCpuTime;processUpTime=newProcessUpTime;currentCpuUsage=Math.max(processCpuUsage,systemCpuUsage);if(currentLoad>SystemRuleManager.getSystemLoadThreshold()){writeSystemStatusLog();}}catch(Throwablee){RecordLog.warn([SystemStatusListener]FailedtogetsystemmetricsfromJMX,e);}}privatevoidwriteSystemStatusLog(){StringBuildersb=newStringBuilder();sb.append(Loadexceedsthethreshold:);sb.append(load:).append(String.format(%.4f,currentLoad)).append(;);sb.append(cpuUsage:).append(String.format(%.4f,currentCpuUsage)).append(;);sb.append(qps:).append(String.format(%.4f,Constants.ENTRY_NODE.passQps())).append(;);sb.append(rt:).append(String.format(%.4f,Constants.ENTRY_NODE.avgRt())).append(;);sb.append(thread:).append(Constants.ENTRY_NODE.curThreadNum()).append(;);sb.append(success:).append(String.format(%.4f,Constants.ENTRY_NODE.successQps())).append(;);sb.append(minRt:).append(String.format(%.2f,Constants.ENTRY_NODE.minRt())).append(;);sb.append(maxSuccess:).append(String.format(%.2f,Constants.ENTRY_NODE.maxSuccessQps())).append(;);RecordLog.info(sb.toString());}}三、京东版最佳实践3.1运用办法Sentinel运用办法自身十分简略，便是一个注解，可是要考虑规矩加载和规矩耐久化的办法，现有的办法有：•运用Sentinel-dashboard功用：运用面板接入需求维护一个装备规矩的办理端，考虑到偏后端的体系需求额定维护一个面板本钱较大，假如是像RPC结构这种自身有办理端的接入能够考虑次计划。

FlowSlot的首要逻辑都在FlowRuleChecker里，华民介绍之前，华民咱们先看一下Sentinel关于规矩的模型描绘，下图分别是限流、拜访操控规矩、体系维护规矩（Linux负载）、降级规矩/***流量操控两种形式*0:threadcount（当调用该api的线程数抵达阈值的时分，进行限流）*1:QPS（当调用该api的QPS抵达阈值的时分，进行限流）*/privateintgrade=RuleConstant.FLOW_GRADE_QPS;/***流量操控阈值，值意义与grade有关*/privatedoublecount;/***调用联系限流战略（能够支撑相关资源或指定链路的多样性限流需求）*直接（api抵达限流条件时，直接限流）*相关（当相关的资源抵达限流阈值时，就限流自己）*链路（只记载指定链路上的流量）*{linkRuleConstant#STRATEGY_DIRECT}fordirectflowcontrol(byorigin);*{linkRuleConstant#STRATEGY_RELATE}forrelevantflowcontrol(withrelevantresource);*{linkRuleConstant#STRATEGY_CHAIN}forchainflowcontrol(byentranceresource).*/privateintstrategy=RuleConstant.STRATEGY_DIRECT;/***Referenceresourceinflowcontrolwithrelevantresourceorcontext.*/privateStringrefResource;/***流控作用：*0.default(rejectdirectly),直接回绝，抛反常FlowException*1.warmup,慢发动形式（依据coldFactor（冷加载因子，默许3）的值，从阈值/coldFactor，经过预热时长，才抵达设置的QPS阈值）*2.ratelimiter排队等候*3.warmup+ratelimiter*/privateintcontrolBehavior=RuleConstant.CONTROL_BEHAVIOR_DEFAULT;privateintwarmUpPeriodSec=10;/***Maxqueueingtimeinratelimiterbehavior.*/privateintmaxQueueingTimeMs=500;/***是否集群限流，默许为否*/privatebooleanclusterMode;/***Flowruleconfigforclustermode.*/privateClusterFlowConfigclusterConfig;/***Thetrafficshaping(throttling)controller.*/privateTrafficShapingControllercontroller;接着咱们持续剖析FlowRuleCheckercanPassCheck第一步会美观limitApp，这个是结合拜访授权约束规矩运用的，默许是一切。

关于数据核算，有精园首要会牵扯到ArrayMetric、BucketLeapArray、MetricBucket、WindowWrap等类。CellularIoTWiki沉积的技能内容方向如下：昆明奇观物联的事务服务范围：根据自研的NB-IoT、Cat1、Cat4等物联网模组，为客户物联网ODM/OEM解决方案服务。

参数：力构力五返回值：力构力无示例：//初始化DI上报定时器dtu_di_times_init();4Demo实战4.1创立一个Demo仿制20.2_at_xtu示例工程，到同一个文件夹下，修正文件名为20.9_di_xtu，如图：4.2修正makefile添加文件组件地点目录头文件途径，和源文件途径，以及一些宏界说，如图：-D是makefile中界说宏界说必要的前缀，能够查找相关makefile学习文章学习相关常识。后来咱们发现，筑中族共许多物联网产品开发团队都面临着类似的困扰，筑中族共所以，咱们决定向整体物联网职业开发者敞开奇观物联内部沉积的运用技能常识库Wiki，希望能为更多物联网产品开发者减轻一些重复造轮子的担负。

从产品界说、华民芯片选型，到软硬件研制和测验，物联网技能的碎片化以及工业资源的碎片化，一向对团队的产品开发交给质量和功率构成限制。状况触发可装备为脉冲计数接口一种上升沿（按键开释计数），有精园一种下降沿（按键闭合计数），第三种电平（按键闭合开释都会计数）。

本文地址：http://hechi.meichubang.com/html/20250305/79.html