今天由我们三位来为大家介绍这一个我们主题不好意思,我们是大会要求用英语,所以...Edge Computing ArchitectAnd then my parents, Roger, and Gandra.Hey!Ok, the IND engineers.We are from Nine Night Cloud Information Technology Company.Nine Night Cloud, a provider of software and service-swing companies.We are front in 2012.Nine Night Cloud focuses on open-soft cloud computing and edge computing.Our enterprise products include the edge cloud and private cloud,Cars, Platform, CMP, and SD1.Ok, for a long time, we made many important code contributionsat the open-soft community.We are a core member of the OpenStack Foundation and CNCFand Open Nightwalk Foundation.In the three latest versions of OpenStack,we are the second contributor in the Rocket version in the world.At the same time, we are the fourth contributor in the Stanley versions.Ok, let me back to the issue.We divide our topics into three parts.First, that is a question,why OpenStack and Kubernetes are better together.To answer this question better,we need to understand the difference betweenOpenStack and Kubernetes.That explains the problem from four SPS.They are computer,nightwalk, storage, and business innovations.Ok, the OpenStack machine uses the hardware visualizationand has a complete color.As you know,the watch machine is isolated,so that the watch machine is secured.But the containers use the operating system visualization,so it provides a lightweight application runtimecompared to OpenStack.Kubernetes has a weightage of agile and possible.But if virtualization technology is the icon to the DNA of cloud computing,OpenStack and Kubernetes has been moved to two direction distant.Ice and pass,you know.Please pay attention,although the container has a namespace,but it can only meet the use of resource.There is no real kernel isolation.Training the isolation depends on the capabilityof the runtime model of the container.Ok,in the past,we were consider about landing a layerof virtualized watch machinebetween hardware and containerto solve these problems.But it is not a good waybecause it is reduced performance of a computer.So the OpenStack foundation has been promotingthe development of the Cata containerto reduce the visualization of a handand why ensuring container security.And on the hand of Lightwork,Kubernetes has many Nightwork modelssuch as ANI,Service,Ingress,and the Nightwork policy.However,the traditional Lightwork departmentbelieved that the Lightwork infrastructureplatform should be have the HCPand supplied DNSto ensure security groupfloating IP and VPC additionally.Building a multi-tell night isolationand scalable SDN Nightworkis very important to traditionallightwork department.So in the great OpenStackKubernetes Lightwork view,Kubernetes is a better choicebecause it can solve the problemof the latest nightoverly Lightworkand achieve the Lightwork connectionbetween the containersand the watch machine.On the hand of the storage,OpenStack can providethree types of storage servicesuch as broad storageand object storageand file storage.To OpenStack,it's importantto use storage resource betterrather than providing storageservice for the clusterbecause the core valueof Kubernetesis always to drivewith these innovations.You know,Continenceis the standard toolfor DevOpsand cloud retailsand the micro-seriesarchitecture is the systemsoftware design ideaof the cloud.Continence change the deliverway of OpenStackof application,sorry.And micro-seriesarchitecture change the formof application.In the end,I think thatin the large-scale cluster,using OpenStackand Kubernetes togetheris a better choice.I believe that in the futureall the applicationswill run in the cloud.But here,I have foundthat in the futureall the applicationswill run in the cloud.But here I havequestionshow to builda uniform monitoringplatform for OpenStackand Kubernetes.I believe Rogeror Rogercan help me to answerthese questions.Welcome Roger.Thanks to my case speech.We first look at how theprom shoes on the OpenStack siteall OpenStack environmentis mainly divided into two types of nodesControl nodesand edge nodesControl node and compute nodeThe self-developed OpenStackexporter collects the metricsrelated to the OpenStackcluster at the control nodeand the collect virtual machineat the compute node.The content of OpenStackexporter will be describedin detail later.Back to our environment,edge node in the OpenStackruns a promise shoes processin order to facilitate theclassification of the collectcenter.The development is further dividedinto central and edge promise shoes.The central promise shoes isonly deployed on the all control nodeto collect content relatedto the OpenStack environmentsuch as indicators of theAZ and HA domains.AZ promise shoes is deployedon any node to collect physicalnode information and metricswith node characteristics.For example,the compute nodecollect metrics relatedto the virtual machine.The periodically pull datafrom the central to the edge nodein a federated manner.The OpenStack environment cancollect the following indicators.We collect basicKPI metrics from physicalnodes and the virtual machineand then aggregate them intothe demands of host aggregateavailability,zoomand resource pool.On the other hand,it will monitor commonfalls in the environment such asnode failure,virtual machinefailure,OpenStack storiesfailure and so on.On the alarmside,an alertmanager cluster is deployedin the OpenStack environment.When an alarm occurs,H node sends an alarm requestto the specifiedan alertmanager clusteraccording to the alarm ruleof its own attributesbecause it isdirect alarm of each node.Let's look at theorange line.The extremely stressed responselink can ensure that criticalalarms can be sent outof course,the reasonfor dare to design thisis that the alertmanageritself has the functionof application filteringnoisy reductionand grouping.Relying on these features avoidsthe risk of criticalalarmsbeing covered by a largenumber of the same typealarms or low purityalarms.Next,let's look atoverall architecture diagram.The monitoringinternal resource indicatoron the Kubernetes sideuse a conventionalpromissure operatorapproach to create promissuresand related resourcethrough the operator andmaintain their life circle.How to achieve unified monitoring.The federal method is still usedwith the central promiseson OpenStack side asthe monitoring point.And the promissures onKubinetic side periodicallyto its scribe metrics.Let's look at thisstudied line.The specific methodis to write the URLto the central promissuresinto the endpoint.Match the service andthe service monitor.And then find and writethe configuration by theoperator.A detailed規模promissures canscalar out to meetthe needs cluster expansionsand can scale upseparate and the monitormetrics.In termsof developmentdeploymentwe initially planto reuse the conceptof promissures operatorand add an external resourcetype.EnablePromissures resourcesdeploy promissures in theOpenStackwhere the docker APRand maintain its life circle.It looks very fashionbut it is actuallynot necessary at all.In the environmentof Kubernetesthe change in the lifeof the pod cause its IP tochange which affectsthe target configurationof promissures.But don't worry about thisin the OpenStack environment.Notes and servicesplanned for IPat thebeginningof development.Deployment andis no longer changes.So the final deploymentis in the form ofanswerabledue tothe variety ofnotes type and thepromissures types.The command configurationcan beautomated during deployment.OpenStackExploreris used to collect resourceindicator such as OpenStackcluster,services,andretro machines.Through a looselycoupled architecture,datafrom multiple layerdifferent types and differentpure wheels are uniformlycollected.It hasfollowing main feature.Lose coupling isinsention and removalof various indicators.Different data acquisitionmodes that allowachieve acquisition andpassive monitoring.Thesignals and asynchronousacquisition modes for anyindicator andare optional.The signalsperiod of anyindicator can be configured.After the project isstart,differentdatatypes areregested into thesignals or asynchronouscontroller according to theconfiguration file during theinitialization process.The types of resourceregest between thecontroller and thecomplementary,and thesignals controller isplantedand thesignals controlleris used to control theacquisition process thatrequires high real-timeperformance.And thesignals controller is usedto control theacquisition process withlarge debt volume.Thesignals acquisition process starts withthe project.Theinitialization processregestor,different typesof thecontroller.Thebackend is called fordata collection duringthe circle.Eachnew circle will overwritethe data from thepureo circle.Thesignalization processis relativelysimple.Thatis when the externalrequest comes in,thesignals acquisition process will be triggered.At this time,the controller will call the corresponding method of the backend for data collection.After thesignalization data iscollected,it will bemerged withsignalsdata for thepureo circle.And finally,themerged data will bereturned.Next,mycollege gardener willgiven a demo.Okay,let'ssee the demo.Thepureo real speaker tellsyou why we needthe Kubernetes.And Ishould prepare anopen-stack environment.Onthis environment,wewill deploy promises,openstack exporter,nodeexporter,and alertmanager.Underthe ansible deploymentand we willthe ansible deploymentwe mighttose ansible deployment havethree special ability.First,automatic,nodeplanning,nodeconfigurationfeels and automaticallylocated by set on thesepability to beprovided by different node when deploying.The third,autodetackenvironment configuration.Okay,andthen Kubernetes environmentshouldservice endpoint and servicemonitor.The service definitionfor externalresource is very similar tonormal service definition,exceptwe are not going towe use port selectors.Wewill create the service anddefine the endpoint ourselves.Thenlet me show a video.Okay,nowlet's see thevideo and I'm going to talkabout the first part,unifiedmonitoring.We need tocollect the open stack datadatause Kubernetes environmentpromises.Firstly,let'slook at which data to be collectedin the open stack environment.Okay,let'ssee the what virtual machinein there.Okay,wecan see there arefeel virtual machine in theenvironment,and next welook at the aggregate.Okay,thethird,we arelook at theavailability zone.Okay,andthen let's take a look at thepromises state in thein the open stack environment.Okay,wecanfind the open stack exportermetrics,likehypervisor info,andVCPU totals.Okay,theintroduce of the open stackenvironment is over.Next,let'ssee the Kubernetes environment.Okay,let us tosee the service.Okay,andthen welook at the endpoint.Okay,nowwe canthis IP isexternal.ThisIP isexternal resource open stackpromises address.Okay,andwe needto emphasize is theaddress of the endpoint wecan see is address ofpromises in the open stackenvironment.Okay,let ussee the promises in Kubernetesenvironment available.Okay,thenwe can see thisdata is the Kubernetesmetrics.Andthen we try to get theopen stack exportermetrics inthe environment,likehypervisor info,andthe CPU totals.Okay,thisdata are the same asbefore wejust know wefound it.Okay,letmeshow theverify the open stackpromises URL isregested.Okay,we cansee the open stackinterface.Wecan find theOSpromises.TheOSpromises ismean thepromise is theexternal resourceopen stackopen stackpromises.Okay,weclink this URLand wewill find the open stackpromises in there.Thenlet us tosee theopen stack databy Grafana fromthe Kubernetesenvironment.First,letme see the virtual machine.Okay,let us to verifyit's virtual machine isguided by open stackenvironment.Okay,andtake a UID in thisplace.Andmyopen stack environment.Okay,wecan get this virtual machinefrom the open stackenvironment.The nextlook at thehost aggregateby Grafana.Okay,andthis place has a CPUand memoryand VMtotals.Thedata is last five minutes.Okay,thenwe look at theValability Zone.Okay,wesee the NOAAValability.Okay,let see the virtual machinesoverview.We have the fileactive virtual machine inopen stack environment.Andthis data isguided by Kubernetespromises.Okay,inthis interface,we cansee the CPU load top threein there.Okay,nowlet me show youthe second part.IfI pre-shortedon the CPU of the host,wherethere is no exporterin the open stack environment,canpromises its Kubernetesenvironment react in time.Okay,wefind a commandto pre-share theCPU.Okay,wait amoment.Okay,we findthis.Okay,one,two,three,four,five,six.Okay,wecan see thisplace had beenchange read.AndtheCPU unitizationread is sorted.Andwe can see thenodeload1.Wecan see thenodeload improve alot from justnow,from thistime.Okay,andthen wefinally let's see ifthe alert managerin the open stackexporter environment hasalert.Look at the alertmanager.Okay,wecan seethe alert managersorry,sorry,sorry.Okay,wecan seethe alert managerwhich is in the open stack environment has alert.Andthis machine isthe hostpre-shared it.Wegot the data veryquickly and make changein the interface.Okay,thankyou.Thank you.Thankyou all.Thank you,DJ.Okay.其實這個監控的問題是比較難處理的,然後用中文都很難說明白,但但會要求我們用英文所以非常的sorry.可是我們現在可以用這個問題,因為相信這個場景我們一直認為是會有很多的一個應用場景的好,謝謝大家,我們已經結束了好,謝謝大家.Thank you.Thank you all.