Cloud Native Ambassador Day - Extending Kubernetes

Extending Kubernetes: Abstract: When somebody asks me why Kubernetes got so much traction the answer is always the same: “its extendibility”. You can run it everywhere, almost all cloud providers or bare metal providers can deliver a managed or integrated distribution of Kubernetes. This is because it acts as a pluggable API gateway to manage your infrastructure across providers. As a developer or infrastructure operator having the ability to write the best integration your company or team needs is a huge plus. During this talk, Gianluca will show you the different integration points you can use to extend Kubernetes such as shared informers, controller, kubectl plugin. Concepts are generically applicable to Kubernetes, code will mainly be in Go.

Transcript

1. Extending Kubernetes The superpower behind the Kubernetes API Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
2. apiVersion: batch/v1 kind: Job metadata: name: hello spec: template: # This is the pod template spec: containers: - name: hello image: busybox command: ['sh', '-c', 'echo "Hello, Kubernetes!" && sleep 3600'] restartPolicy: OnFailure # The pod template ends here Example coming from kubernetes.io Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
3. Kubernetes is a declarative framework If you use it just as end tool you are missing its real value
4. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
5. k-proxy kubelet sched sched sched Control Plane Node etcd Kubernetes cluster api api api c-c-m c-c-m c-c-m c-m c-m c-m Node Node k-proxy kubelet kubelet k-proxy Control plane Scheduler sched Cloud controller manager (optional) c-c-m Controller manager c-m kubelet kubelet kube-proxy k-proxy (persistence store) etcd etcd Node API server api from kubernetes.io Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
6. from SlideShare: Moby CRI Containerd Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
7. How you justify Kubernetes Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
8. Play your own game. That's why services have their own API. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
9. You know yourself, your team and your product. Build around your requirement Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
10. Who am I? Software Engineer at Equinix Metal (Packet) Open Source maintainer for Kubernetes, Docker, TestContainer Docker Captain and CNCF Ambassador When not coding I grow vegetables I am active on Twitter as @gianarb
11. Kubernetes follows the same rule, if kubectl is not enough because you have an half way workflow that you like, or the kubectl makes your solution hard to maintain, it is OK to build something by yourself. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
12. There different ways to extend Kubernetes, here a couple: . A kubectl plugin. . Via Client-GO or any other SDK. . Custom Resource Definition. . Aggregation layer Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
13. Let's start with kubectl . It is by far the most flexible and easy way to extend Kubernetes. You have to deliver an executable in your $PATH that starts with kubectl-* . Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
14. Examples: A binary called kubectl-ns can be executed as kubectl ns A binary called kubectl-profefe can be executed as kubectl profefe Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
15. kubectl-profefe Profefe is an open source project to do continuous profiling of application using pprof, such as Golang. cron job application runtime/pprof net/http/pprof pull Storage developer profefe-collector query push GET /debug/pprof/profile Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
16. I wrote a project called profefe/kube-profefe that acts as a bridge between profefe and Kubernetes. . It serves a binary called kprofefe , it can run as Kubernetes cronjob and it collects profiles targeting application running in Kubernetes. . kubectl-profefe helps you to interact with profefe and Kubernetes. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
17. $ kubectl profefe --help It is a kubectl plugin that you can use to retrieve and manage profiles in Go. Available Commands: capture Capture gathers profiles for a pod or a set of them. If can filter by namespace and via label selector. get Display one or many resources help Help about any command load Load a profile you have locally to profefe Flags: -A, --all-namespaces If present, list the requested object(s) across all namespaces --as string Username to impersonate for the operation --as-group stringArray Group to impersonate for the operation, this flag can be repeated to specify multiple groups. --cluster string The name of the kubeconfig cluster to use --context string The name of the kubeconfig context to use -f, --filename strings identifying the resource. -h, --help help for kubectl-profefe --insecure-skip-tls-verify If true, the server's certificate will not be checked for validity. This will make your HTTPS connections insecure --kubeconfig string Path to the kubeconfig file to use for CLI requests. -n, --namespace string If present, the namespace scope for this CLI request -l, --selector string Selector (label query) to filter on, supports '=', '==', and '!='.(e.g. -l key1=value1,key2=value2) Use "kubectl-profefe [command] --help" for more information about a command. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
18. Links about kubectl plugins: "My experience with Krew to manage kubectl plugins" "kubectl flags in your plugin" Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
19. Client GO and other SDKs Kubernetes API works as any other API. There are client libraries that you can use in many languages or it works via HTTP, and all the languages have an HTTP client available. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
20. Not all the languages are the same There are official supported an unofficial libraries available out there. Go, Javascript, Haskell, Python, Java, Dotnet are supported by the Kubernetes community, you can check out the actual documentation. All of them are in different state, Golang is well done. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
21. kprofefe the binary I told you about before uses the client-go library to do a couple of things: Retrieve pods filtered according to label section or/and per namespace in order to get the right targets for profefe It uses the annotations for a particular pod in order to figure out where the pprof server runs (the right port and path) Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
22. Get pods // GetSelectedPods returns all the pods with the profefe annotation enabled // filtered by the selected labels func GetSelectedPods(clientset kubernetes.Interface, namespace string, listOpt metav1.ListOptions) ([]v1.Pod, error) { target := []v1.Pod{} pods, err := clientset.CoreV1().Pods(namespace).List(listOpt) if err != nil { return target, err } for _, pod := range pods.Items { enabled, ok := pod.Annotations[ProfefeEnabledAnnotation] if ok && enabled == "true" && pod.Status.Phase == v1.PodRunning { target = append(target, pod) } } return target, nil } Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
23. Port forward programmatically func PortForwardAPod(req PortForwardAPodRequest) error { path := fmt.Sprintf("/api/v1/namespaces/%s/pods/%s/portforward", req.Pod.Namespace, req.Pod.Name) hostIP := strings.TrimLeft(req.RestConfig.Host, "htps:/") transport, upgrader, err := spdy.RoundTripperFor(req.RestConfig) if err != nil { return err } dialer := spdy.NewDialer( upgrader, &http.Client{Transport: transport}, http.MethodPost, &url.URL{Scheme: "https", Path: path, Host: hostIP}) fw, err := portforward.New( dialer, []string{fmt.Sprintf("%d:%d", req.LocalPort, req.PodPort)}, req.StopCh, req.ReadyCh, req.Streams.Out, req.Streams.ErrOut) if err != nil { return err } return fw.ForwardPorts() } Repository: gianarb/kube-port-forward
24. Custom Resource Definition (CRD) Terraform has the concept of modules. A Terraform module is a bridge between an external resource and Terraform. A CRD is the same, but with Kubernetes. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
25. Custom Resource Definition (CRD) Native resources: Pod, Services, Deployment, StatefulSet, Ingress... Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
26. AWS Controllers for Kubernetes: Elastic Load Balancer, S3 bucket, CloudFormation Stack. Checkout: ACK Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
28. client api-se kube crd-cont Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
29. Why you should write a CRD? Your team uses Kubernetes a lot, they know how to use the kubectl and client-go. You can make available for them external services like DNS management or a binary/image repository with an user experience they already know. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
30. Kubernetes gives you a bunch of useful features that you can leverage: . Authentication . Authorization . Audit Logs . Event System . CLI (kubectl) and client libraries Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
31. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb
32. Thank you Any question? Reach out to me via Twitter @gianarb. Gianluca Arbezzano - gianarb.it - twitter.com/gianarb