Key	take	aways
”O	que	nós temos que	aprender?”
https://goo.gl/GP9E8t
Agenda
1. KubeCon – sobre o evento
2. Atualização de projetos CNCF
3. Netflix Keynote – Continuous Delivery
4. HBO Kubernetes Journey
5. AWS Kubernetes Keynote
6. Nordstrom 101 Ways to Crash your Cluster
7. Cultura Open Source à Google
8. Kubernetes em 2018
9. Links
”Exciting Time for Boring Infrastructure”
0
500
1000
1500
2000
2500
3000
3500
4000
4500
2014 2015	San	Francisco 2016	Seattle 2017	Austin
Crescimento	do	interesse	em	Kubernetes
Participantes
”Exciting Time for Boring Infrastructure”
Quem é CNCF?
Kubernetes
Orchestration
Prometheus
Monitoring
OpenTracing
Dist.	Tracing API
Fluentd
Logging
Linkerd
Service	Mesh
gRPC
Remote	Procedure
Call
CoreDNS
Service
Discovery
Containerd
Container
Runtime
CNI
Networking	API
Envoy
Service	Mesh
Notary
Security
TUF
Software	Update
CLOUD NATIVE LANDSCAPE
Atualizações dos projetos CNCF
Michelle Noorali
Software Engineer,
Microsoft Azure
Kubernetes 1.9
Suporte	pra	Windows	
Containers
CoreDNS substitui	o	
KubeDNS
Linkerd
Atualmente	com	50	
usuários	em	produção
Potencial	substituto	
à Conduit
Containerd 1.0.0
Prometheus 2.0
OpenTracing à Jaeger
Security	à TUF
GA	à produção
Melhoria	na	performance
Nascido	no	Uber
Netflix keynote
• Jornada para Enterprise Continuous Delivery
• Desenvolvimento interno de ferramentas à projeto open source
• Adotado pela Netflix e pelo Waze
https://www.slideshare.net/diannemarsh/kubeconcloud-native-keynote-december-2017
Multi Cloud Releases	Automatizados Melhores	práticas	de	deploy ”built in” Comunidade	ativa
HBO – traffic pattern
Problemas
• Desperdicio:
• Nodejs + EC2 = +50% CPU não
utilizado
• Multiplicado por overscaling
• Slow to scale up/down
• Service to service à ELB everywhere
Why Kubernetes?
• Utilization
• Introspection
• Faster
• Deployments and Rollbacks
• Pod Autoscaling
• Safer
• Rolling Updates
• Configuration as code
• Batteries are included
• DNS, SD, kubectl
Step 1
• End of 2015
• 0 containerized services
• July 2017
• GoT Season 7 Premiere
• Dockerize
• Modify EC2 Deployments
• Modify CI/CD Pipelines
Step Next
• Run Kubernetes
• But...
AWS Cloud Native Keynote
Adrian Cockcroft,
VP Cloud Architecture Strategy
https://youtu.be/5U-6sxR5DaQ
Cloud Native Principles
• Pay as you go, afterwards
• Self service – no waiting
• Global distributed by default
• Cross region availability models
• High utilization – turn idle resourcess off
• Immutable code deployments
101 Ways to Crash Your Cluster
Marius Grigoriu, Sr. Technology Manager, Nordstrom
Emmanuel Gomez, Principal Engineer, Nordstrom
Kubernetes @Nordstrom
• Microserviços
• Reviews
• Giftcard
• Purchase Orders
• Authentication
• Personalization
• And more
• Dev Tools
• Issue tracking
• Build runners
• Log aggregation
• Telemetry aggregation
• Alerting
Nordstrom: 101 ways to crash your cluster
• Caso 1: Os nós (workers) que paravam
• Nós que deixaram de responder 25% à 50%
• Todos os nós se recuperavam ao mesmo tempo
• 15 minutos – kubelets deixa de responder
• Caused by lack of timeout or heartbeat kubelet->apiserver
• Fixed in 1.8 (backported to 1.7.8)
• Caso 2: o dia que os robôs se viraram contra eles
• Sem aviso o cluster é encolhido para alguns poucos nós
• Nunca foi possível determinar a causa raíz pois os dados de
diagnóstico se perderam
• Caso 3: o etcd com múltiplas personalidades
• As informações deixaram de fazer sentido
• kubectl get pods.. retornava dois resultados diferentes, de forma
aleatória
• Impacto: queda completa do cluster
• Causa: problema na consistência do etcd provavelmente causado por
latência de disco em um ambiente de HA à resolvido kubernetes 1.9
NotReady Checklist
1. Run: kubectl describe node
2. “kubelet stopped posting node status”
3. Look for signs of high resource utilization
4. Is there a networking issue?
5. kubelets messedup?
6. apiserver messedup?
7. Oh look, everything’s OK now
Cultura Open Source à Google
• Sarah Novotny, Head of Open Source Strategy, Google
• ”Culture eats strategy for breakfast” Peter Druker
https://www.youtube.com/watch?v=-5R_GbGg1nI
Exclusivo INCLUSIVO
Centralizado DISTRIBUIDO
Empresa COMUNIDADE
Processo AUTOMAÇÃO
Estagnação CRESCIMENTO
Valores
Futuro do Kubernetes - 2018
Build faster, smarter, better – the year of service mesh
1. Make Microservices Easier
Istio, Envoy, Conduit
2. Data for Everyone Make Data Workloads Easier
TensorFlow, GPU support, Spark
3. Integrate Servless Natively
openwhisk, fission, kubeless
4. Improve App Deployment & Config
HELM 3.0, kompose
5. Extensible and Secure Workload Identity
Spiffe, kerberos, container-identity-wg
6. Policy, Multi-tenancy, integration
Open policy agent, LDAP, etc.
7. Better Containers (and VMs)
oci, containerd, kubevirt, hyperv
Clayton Coleman
Architect, Kubernetes
and OpenShift, Red Hat
Kubernetes = futuros empregos TI
Links
• Agenda e apresentações
• https://kccncna17.sched.com/
• Playlist com os vídeos das sessões
• https://goo.gl/GP9E8t

Kubecon 2017 Resumo

  • 1.
  • 2.
    Agenda 1. KubeCon –sobre o evento 2. Atualização de projetos CNCF 3. Netflix Keynote – Continuous Delivery 4. HBO Kubernetes Journey 5. AWS Kubernetes Keynote 6. Nordstrom 101 Ways to Crash your Cluster 7. Cultura Open Source à Google 8. Kubernetes em 2018 9. Links
  • 3.
    ”Exciting Time forBoring Infrastructure” 0 500 1000 1500 2000 2500 3000 3500 4000 4500 2014 2015 San Francisco 2016 Seattle 2017 Austin Crescimento do interesse em Kubernetes Participantes
  • 4.
    ”Exciting Time forBoring Infrastructure”
  • 5.
    Quem é CNCF? Kubernetes Orchestration Prometheus Monitoring OpenTracing Dist. TracingAPI Fluentd Logging Linkerd Service Mesh gRPC Remote Procedure Call CoreDNS Service Discovery Containerd Container Runtime CNI Networking API Envoy Service Mesh Notary Security TUF Software Update
  • 6.
  • 7.
    Atualizações dos projetosCNCF Michelle Noorali Software Engineer, Microsoft Azure Kubernetes 1.9 Suporte pra Windows Containers CoreDNS substitui o KubeDNS Linkerd Atualmente com 50 usuários em produção Potencial substituto à Conduit Containerd 1.0.0 Prometheus 2.0 OpenTracing à Jaeger Security à TUF GA à produção Melhoria na performance Nascido no Uber
  • 8.
    Netflix keynote • Jornadapara Enterprise Continuous Delivery • Desenvolvimento interno de ferramentas à projeto open source • Adotado pela Netflix e pelo Waze https://www.slideshare.net/diannemarsh/kubeconcloud-native-keynote-december-2017 Multi Cloud Releases Automatizados Melhores práticas de deploy ”built in” Comunidade ativa
  • 9.
    HBO – trafficpattern Problemas • Desperdicio: • Nodejs + EC2 = +50% CPU não utilizado • Multiplicado por overscaling • Slow to scale up/down • Service to service à ELB everywhere
  • 10.
    Why Kubernetes? • Utilization •Introspection • Faster • Deployments and Rollbacks • Pod Autoscaling • Safer • Rolling Updates • Configuration as code • Batteries are included • DNS, SD, kubectl
  • 11.
    Step 1 • Endof 2015 • 0 containerized services • July 2017 • GoT Season 7 Premiere • Dockerize • Modify EC2 Deployments • Modify CI/CD Pipelines
  • 12.
    Step Next • RunKubernetes • But...
  • 13.
    AWS Cloud NativeKeynote Adrian Cockcroft, VP Cloud Architecture Strategy https://youtu.be/5U-6sxR5DaQ Cloud Native Principles • Pay as you go, afterwards • Self service – no waiting • Global distributed by default • Cross region availability models • High utilization – turn idle resourcess off • Immutable code deployments
  • 14.
    101 Ways toCrash Your Cluster Marius Grigoriu, Sr. Technology Manager, Nordstrom Emmanuel Gomez, Principal Engineer, Nordstrom
  • 15.
    Kubernetes @Nordstrom • Microserviços •Reviews • Giftcard • Purchase Orders • Authentication • Personalization • And more • Dev Tools • Issue tracking • Build runners • Log aggregation • Telemetry aggregation • Alerting
  • 16.
    Nordstrom: 101 waysto crash your cluster • Caso 1: Os nós (workers) que paravam • Nós que deixaram de responder 25% à 50% • Todos os nós se recuperavam ao mesmo tempo • 15 minutos – kubelets deixa de responder • Caused by lack of timeout or heartbeat kubelet->apiserver • Fixed in 1.8 (backported to 1.7.8) • Caso 2: o dia que os robôs se viraram contra eles • Sem aviso o cluster é encolhido para alguns poucos nós • Nunca foi possível determinar a causa raíz pois os dados de diagnóstico se perderam • Caso 3: o etcd com múltiplas personalidades • As informações deixaram de fazer sentido • kubectl get pods.. retornava dois resultados diferentes, de forma aleatória • Impacto: queda completa do cluster • Causa: problema na consistência do etcd provavelmente causado por latência de disco em um ambiente de HA à resolvido kubernetes 1.9
  • 17.
    NotReady Checklist 1. Run:kubectl describe node 2. “kubelet stopped posting node status” 3. Look for signs of high resource utilization 4. Is there a networking issue? 5. kubelets messedup? 6. apiserver messedup? 7. Oh look, everything’s OK now
  • 18.
    Cultura Open Sourceà Google • Sarah Novotny, Head of Open Source Strategy, Google • ”Culture eats strategy for breakfast” Peter Druker https://www.youtube.com/watch?v=-5R_GbGg1nI Exclusivo INCLUSIVO Centralizado DISTRIBUIDO Empresa COMUNIDADE Processo AUTOMAÇÃO Estagnação CRESCIMENTO Valores
  • 19.
    Futuro do Kubernetes- 2018 Build faster, smarter, better – the year of service mesh 1. Make Microservices Easier Istio, Envoy, Conduit 2. Data for Everyone Make Data Workloads Easier TensorFlow, GPU support, Spark 3. Integrate Servless Natively openwhisk, fission, kubeless 4. Improve App Deployment & Config HELM 3.0, kompose 5. Extensible and Secure Workload Identity Spiffe, kerberos, container-identity-wg 6. Policy, Multi-tenancy, integration Open policy agent, LDAP, etc. 7. Better Containers (and VMs) oci, containerd, kubevirt, hyperv Clayton Coleman Architect, Kubernetes and OpenShift, Red Hat
  • 20.
  • 21.
    Links • Agenda eapresentações • https://kccncna17.sched.com/ • Playlist com os vídeos das sessões • https://goo.gl/GP9E8t