Jaeger开源平台技术深入介绍

JaegerProjectDeepDiveJaeger开源平台技术深入介绍ProjectNewFeaturesRoadmapQ&AAgendaWhatisTracing&Why?ConceptsandterminologyModernDistributedSystemsareCOMPLEXLoadingNetflixorFacebookhomepage⇒dozensofmicroservices,100sofnodes6BILLIONSoftimesaday!Howcanwetellwhatisgoingon?Whichserviceistoblamewhenthingsgowrongorbecomeslow?Traditionalmonitoringtoolsdon’thelpMetricsshowsomethingiswrong,butdonotexplainwhy.Logsareamess:concurrentrequests,multiplehosts,impossibletocorrelate.Monitoringtoolsmusttellstories!Doyoulikedebuggingwithoutastacktrace?Weneedtomonitordistributedtransactions⇒distributedtracing!InspiredbyGoogle’sDapperandOpenZipkinStartedatUberinAugust2015OpensourcedinApril2017JoinedCNCFinSep2017(incubating)Applyingforgraduationhttps://github.com/cncf/toc/pull/171Jaeger-/ˈyāɡər/,noun:hunterJaeger,aDistributedTracingPlatformtracecollectionbackendhttps://jaegertracing.iovisualizationfrontendinstrumentationlibrariesdataminingplatformJaegerArchitecture(v2)HostorContainerApplicationjaeger-clientjaeger-agentSpans(UDP)ControlflowjaegercollectorControlflowpoll(sampling,etc.)DBadaptivesamplingjaegerqueryFlinkstreamingKafkapushjaegeringester&indexer asyncUIjaegercollectorJaegerArchitecture(v2)HostorContainerjaeger-clientjaeger-agentopencensus-collectorDBjaegerqueryFlinkstreamingKafkapushjaegeringester&indexer asyncopencensus-agentUIApplicationzipkin-clientSpans(UDP)JaegerArchitecture(v2)HostorContainerjaeger-clientjaeger-agentDBjaegerquerypopencensus-agentUIApplicationzipkin-clientSpans(UDP)opencensus-collectorushJaegerArchitecture(v2)HostorContainerApplicationjaeger-clientjaeger-agentSpans(UDP)ControlflowjaegercollectorControlflowpoll(sampling,etc.)DBadaptivesamplingjaegerqueryFlinkstreamingKafkapushjaegeringester&indexer asyncUIJaeger<3OpenStandards+ =OpenCensushttps://medium.com/jaegertracing/jaeger-and-opentelemetry-1846f701d9f2TechnologyStackGobackendPluggablestorageCassandra,Elasticsearch,memory,...React/JavascriptfrontendOpenTracingInstrumentationlibrariesIntegrationwithKafka,ApacheFlink17ApacheCassandra®isatrademarkoftheApacheSoftwareFoundationintheUnitedStatesand/orothercountries.Project&Community7maintainers,fromUberandRedHatGitHubstats>8,200stars,>810forks>580contributors>220authorsofcommitsandpullrequests>350issuecreatorsJaeger1.10-1.12NewFeaturesNewFeaturesElasticsearchimprovements(security,FindTraceIDs,archiving)BettergRPCsupport(security,retry,loadbalancing,externalresolvers)BetterZipkincompatibilityUIimprovements(tracedetail,find,querycapabilities)https://github.com/jaegertracing/jaeger/releaseshttps://github.com/jaegertracing/jaeger/blob/master/CHANGELOG.mdhttps://github.com/jaegertracing/jaeger-ui/blob/master/CHANGELOG.mdWebsite(easytocontribute)Demo!GraphVisualizationsTradeDiffsandTraceGraphGraphVisualizationsGanttchartisnotgreatfortraceswith10softhousandsofspansTraceDiffsComparetwotracesCompareonetraceagainstagroupoftraces(comingsoon)TraceGraph(comingsoon)Callgraphvisualizationwithmini-aggregationsShowingpathsratherthanindividualRPCsComparingtracestructures–UnifieddiffComparingtracestructures–SharedstructureComparingtracestructures–AbsentinoneofthetracesComparingtracestructures–MoreorlesswithinanodeComparingtracestructures–Substantialdivergence"Youhaveanoutstandingbalance…"Structuralvs.TimeStructuralvs.Time–VerysimilarstructuresStructuralvs.Time–2.74secondsStructuralvs.Time–50%increaseindurationStructuralvs.Time–Arethesenewspanstoblame?Structuralvs.Time–Oristhelagincreasedthroughout?Comparingspandurations–ComingsoonComparingspandurations–SimilardurationsComparingspandurations–Nodesthataren'tsharedComparingspandurations–FollowtheslowernodesComparingspandurations–Comingsoon...Comparingspandurations–Comingsoon...IntegrationsIntegrationsJaegerOperatorforKuberneteshttps://github.com/jaegertracing/jaeger-operatorOpenCensuslibrariesandagentshipwithreceivers/exportersforJaegerhttps://opencensus.io/guides/exporters/supported-exporters/java/jaeger/IstiocomeswithJaegerincludedhttps://istio.io/docs/tasks/telemetry/distributed-tracing/EnvoyworkswithJaegernativeC++clienthttps://www.envoyproxy.io/docs/envoy/latest/start/sandboxes/jaeger_native_tracingEclipseTraceCompassincubatorsupportsimportingJaegertraceshttps://github.com/tuxology/tracevizlab/tree/master/labs/303-jaeger-opentracing-tracesLearnMoreWebsite:jaegertracing.io/Blog:medium.com/jaegertracingGettinginTouchGitHub:https://github.com/jaegertracingChat:https://gitter.im/jaegertracing/MailingList-jaeger-tracing@googlegroups.comBlog:https://medium.com/jaegertracingTwitter:https://twitter.com/JaegerTracingBi-WeeklyCommunityMeetingsQ&AOpenDiscussionAppendixAdditionalTopicsandSlidesArchitectureChangesAsynchronousIngestionArchitecture2017:PushHostorContainerjaeger-agentjaeger-collectorControlflowpoll(sampling,etc.)DBadaptivesamplingjaeger-querySparkjobspushApplicationjaeger-clientSpans Control(UDP) flowUIAsynchronousspaningestionPushmodelwasstrugglingtokeepupwithtrafficspikesBecauseofsyncstoragewritesCollectorshadtodropdatarandomlyKafkaismuchmoreelasticforwritesJustrawbytes,noschema,noindexingAlotlessoverheadonthewritepathDatainKafkaallowsforstreamingdatamining&aggregationsTwonewcomponents:jaeger-ingesterandjaeger-indexerArchitecturenow:Push+Async+StreamingHostorContainerApplicationjaeger-clientjaeger-agentSpans(UDP)ControlflowjaegercollectorControlflowpoll(sampling,etc.)DBadaptivesamplingjaegerqueryFlinkstreamingKafkapushjaegeringester&indexer asyncUIProtobuf&gRPCEnablingroadmapProtobuf&gRPCInternaldatamodelgeneratedfromProtobufIDLgRPCconnectionbetweenjaeger-agentandjaeger-collectorWhygRPCplaysbetterwithmodernroutingthanTChannelPathtoofficialdatamodelandcollector/queryAPIsProtobuf-basedJSONAPIUnblockdevelopmentofstorageplugins(Thriftstillsupportedforbackwardscompatibility)ZipkinCompatibilityZipkinCompatibilityClientsZipkinB3-***headersforcontextpropagationInteropbetweenJaeger-instrumentedandZipkin-instrumentedappsCollectorZipkinThriftandJSONv2spanformatUseZipkininstrumentation(e.g.Brave)tosendtracestoJaegerOutstandingAcceptZipkinspansfromKafkastreamRoadmaphttp://bit.do/jaeger-roadmapAdaptiveSamplingProblemAPIshaveendpointswithdifferentQPSServiceownersdonotknowthefullimpactofsamplingprobabilityAdaptiveSamplingisperservice+endpoint,decidedbyJaegerbackendbasedontrafficJaegerclientssupportperservice/endpointsamplingstrategiesCanbestaticallyconfiguredincollectorPullrequestsfordynamicrecalculationsAdaptiveSamplingStatusBasedonKafkaandApacheFlinkSupportaggregationsanddataminingExamples:PairwisedependenciesdiagramPath-baseddependenciesdiagramLatencyhistogramsDataPipelineBasedongRPC/ProtobufworkPRsinprogressforproofofconceptCommunitysupportfordifferentstoragebackendsStoragepluginsAddabilitytostore/retrievepartialspansUsecase:Certainworkflowsarehourslong.Unfortunatelyspansareonlyemittedonceafterit’sFinished().“Rootspan”ismissinguntilthecompleteworkflowisfinished.PartialSpans(communitydriven)Thanks