kubeadm Cluster Creation Internals: From Self-Hosting to Upgradability and HA

This talk was given at KubeCon North America 2017 in Austin

Can be viewed online here: https://docs.google.com/presentation/d/1Gp-2blk5WExI_QR59EUZdwfO2BWLJqa626mK2ej-huo/edit#slide=id.p
Description: https://sched.co/CU7I
Recording: https://youtu.be/YCOWQIFVAbg
Location: Austin Convention Center, Austin, TX

Transcript

1. kubeadm Cluster Creation Internals From Self-Hosting to Upgradability and HA Lucas Käldström 7th December 2017 - KubeCon Austin
2. $ whoami Lucas Käldström, Upper Secondary School Student, just turned 18 CNCF Ambassador and Certified Kubernetes Administrator Speaker at KubeCon Berlin 2017 and now here at KubeCon Austin Kubernetes Maintainer since April 2016, active in the community for +2 years SIG Cluster Lifecycle co-lead and kubeadm maintainer Driving luxas labs which currently performs contracting for Weaveworks A guy that has never attended a computing class Image credit: Dan Kohn
3. Agenda 1. What kubeadm is & how it fits into the ecosystem 2. What’s common for every TLS-secured Kubernetes cluster 3. What self-hosting is & how you can self-host your cluster 4. How kubeadm handles upgrades 5. How high availability can be achieved with kubeadm This talk will dive deep into the technical aspects of creating Kubernetes clusters
4. What is kubeadm and why should I care? = A tool that sets up a minimum viable, best-practice Kubernetes cluster Master A Master N* Node 1 Node N kubeadm kubeadm kubeadm kubeadm Cloud Provider Load Balancers Monitoring Logging Cluster API Spec Cluster API Cluster API Implementation Addons API* Kubernetes API Bootstrapping Machines Infrastructure Layer 2 Layer 3 Layer 1 *=Yet to be done/WIP
5. kops kubeadm vs kops Two different projects, two different scopes Master A Master N* Node 1 Node N kubeadm kubeadm kubeadm kubeadm Cloud Provider Load Balancers Monitoring Logging Cluster API Spec Cluster API Implementation Addons API* Kubernetes API Bootstrapping Machines Infrastructure Cluster API *=Yet to be done/WIP
6. Key design takeaways - kubeadm’s task is to set up a best-practice cluster for each minor version - The user experience should be simple, and the cluster reasonably secure - kubeadm’s scope is limited; intended to be a building block - Only ever deals with the local filesystem and the Kubernetes API - Agnostic to how exactly the kubelet is run - Setting up or favoring a specific CNI network is out of scope - Composable architecture with everything divided into phases Audience: build-your-first-own-cluster users & higher-level tools like kops & kubicorn
7. Nodes Master Kubernetes’ high-level component architecture Node 3 OS Container Runtime Kubelet Networking Node 2 OS Container Runtime Kubelet Networking Node 1 OS Container Runtime Kubelet Networking API Server (REST API) Controller Manager (Controller Loops) Scheduler (Bind Pod to Node) etcd (key-value DB, SSOT) User Legend: CNI CRI OCI Protobuf gRPC JSON
8. What does `kubeadm init` really do -- part 1? First kubeadm creates the necessary certificates for setting up a cluster with TLS-secured communication and API Aggregation support. Then client certificates are generated in KubeConfig files for the first actors that need identities Phase: Certificates (/etc/kubernetes/pki) API Serving Cert Root CA Cert API Kubelet Clientcert Front Proxy CA Cert ServiceAccount Private Key API Clientcert Phase: Kubeconfig (/etc/kubernetes/*.conf) Kubelet Clientcert Root CA Cert Admin Clientcert K-C-M Clientcert Scheduler Clientcert Needed for all clusters kubeadm-specific
9. What does `kubeadm init` really do -- part 2? kubeadm generates Static Pod manifest files for etcd (or use external etcd) and control plane components kubelet runs the Static Pods and hence kubeadm assumes a running kubelet kubeadm waits for the kubelet to start the Static Pods, an operation that may fail in many ways Phase: Etcd (/etc/kubernetes/manifests) Local etcd Static Pod, localhost:2379 The running kubelet starts the Static Pods Phase: Control Plane (/etc/kubernetes/manifests) API Server, ${MASTER_IP}:6443 Controller Manager, localhost:10251 Scheduler, localhost:10252 Host etcd externally Needed for all clusters kubeadm-specific
10. What does `kubeadm init` really do -- part 3? In order to make the master exclusive and identifiable, it is tainted and labelled with a common-practice label For `kubeadm upgrade` to remember the config passed to `kubeadm init`, the config is uploaded to the cluster A Node Bootstrap Token is created and granted privileges to add a node Lastly, kube-proxy and kube-dns / CoreDNS are deployed Phase: Mark the master node kubectl taint node <master> node-role.kubernetes.io/master=”” Phase: Upload kubeadm Configuration kubectl label node <master> node-role.kubernetes.io/master=”” kubectl -n kube-system create configmap kubeadm-config --from-file kubeadm.yaml Phase: Create a Bootstrap Token for a node kubeadm token create <token> Phase: Deploy mandatory add-ons kube-proxy DaemonSet kube-dns / CoreDNS Deployment Needed for all clusters kubeadm-specific
11. Build your cluster from scratch with `kubeadm phase` The `kubeadm phase` command lets you invoke atomic sub-tasks of a full installation operation. You don’t have to choose the “full meal deal” anymore (kubeadm init). This is what `kubeadm init` looks like phase-wise: $ kubeadm alpha phase certificates all $ kubeadm alpha phase kubeconfig all $ kubeadm alpha phase etcd local $ kubeadm alpha phase controlplane all $ systemctl start kubelet $ kubeadm config upload from-flags $ kubeadm alpha phase mark-master $(kubectl get no --no-headers | awk ‘{print $1}’) $ kubeadm alpha phase bootstrap-token cluster-info /etc/kubernetes/admin.conf $ kubeadm alpha phase bootstrap-token node allow-post-csrs $ kubeadm alpha phase bootstrap-token node allow-auto-approve $ kubeadm token create $ kubeadm alpha phase addons kube-dns $ kubeadm alpha phase addons kube-proxy
12. Setting up a dynamic TLS-secured cluster Nodes Master API Server Controller Manager Scheduler CN=system:kube-controller-manager CN=system:kube-scheduler Kubelet: node-1 HTTPS (6443) Kubelet client O=system:masters Self-signed HTTPS (10250) CN=system:node:node-1 O=system:nodes Kubelet: node-2 (to be joined) Self-signed HTTPS (10250) Bootstrap Token & trusted CA CN=system:node:node-2 O=system:nodes CSR Approver CSR Signer Legend: Logs / Exec calls Normal HTTPS POST CSR SAR Webhook PATCH CSR node-1 CSR node-2 CSR Bootstrap Token CSR=Certificate Signing Request, SAR=Subject Access Review
13. kubeadm & self-hosted Kubernetes cluster Self-hosting =using Kubernetes primitives (e.g DaemonSets, ConfigMaps) to run and configure the control plane itself The self-hosting concept was initially developed by CoreOS and the bootkube team. Also see the other self-hosting talk here at KubeCon. We’re now in the process of “upstreaming” the work done in bootkube, in a way that makes it easier for any Kubernetes cluster to be self-hosted. Building blocks that can be used for pivoting to self-hosting exist in kubeadm
14. How is a self-hosted cluster bootstrapped? kubeadm either hosts the control plane in Static Pods managed in the filesystem or as self-hosted DaemonSets. The process is modular; it creates the self-hosted control plane from the current state of the world of Static Pods.
15. Upgrading clusters with kubeadm 1. Upgrades made easy with `kubeadm upgrade plan` and `kubeadm upgrade apply` 2. In v1.8, the upgrades basically shifted Static Pod files around on disk 3. Self-hosted upgrades in HA clusters work by modifying normal Kubernetes resources 4. From v1.9 we’re supporting automated downgrades as well 5. In a future release we’ll look at “runtime reconfiguration” i.e. invoking an upgrade with new config but the same version number 6. Read the proposal
16. The “just works” kubeadm HA feature request Adding support for joining masters is the most popular feature request What challenges do we face in making a slick multi-master UX like this? Three primary challenges: - Managing state reliably in an HA etcd cluster with TLS communication - Sharing & rotating the common certificates from master A to master B - Making the kubelets able to address all masters & dynamic reconfig $ kubeadm init … - Add node: kubeadm join --token abcdef.0123456789abcdef 192.168.1.100:6443 - Add master: kubeadm join master --token fedcba.fedcba09876543210 192.168.1.100:6443
17. HA etcd cluster External Load Balancer or DNS-based API server resolving How achieve HA with kubeadm today? Master A (kubeadm init) API Server Controller Manager Scheduler Shared certificates etcd etcd etcd Master B (kubeadm init) API Server Controller Manager Scheduler Shared certificates Master C (kubeadm init) API Server Controller Manager Scheduler Shared certificates Nodes (kubeadm join) Kubelet 1 Kubelet 2 Kubelet 3 Kubelet 4 Kubelet 5 Do-it-yourself 1. Set up HA etcd cluster 2. Copy certificates from master A to B and C 3. Set up a loadbalancer in front of the API servers
18. Master A, kubeadm init API Server, DS Controller Manager, DS Scheduler, DS Certificates, Secrets Kubelet A etcd-operator, Deploy. etcd A, Pod Master B, kubeadm join master API Server, DS Controller Manager, DS Scheduler, DS Certificates, Secrets Kubelet B etcd B, Pod Master C, kubeadm join master API Server, DS Controller Manager, DS Scheduler, DS Certificates, Secrets Kubelet C etcd C, Pod Node 2, kubeadm join Kubelet 2 kube-proxy, DS Envoy, DS? Node 1, kubeadm join Kubelet 1 kube-proxy, DS Envoy, DS? Node 3, kubeadm join Kubelet 3 kube-proxy, DS Envoy, DS? One possible (?) solution Read the full proposal document DS=DaemonSet
19. What now? Follow the SIG Cluster Lifecycle YouTube playlist Check out the meeting notes for our weekly SIG meetings in Zoom Join the #sig-cluster-lifecycle (for dev) and #kubeadm (for support) Check out Diego Pontoriero’ talk “Self-Hosted Kubernetes: How and Why” Read the two latest SIG updates on the Kubernetes blog in January and August Check out the kubeadm setup guide, reference doc and design doc Read what the kubeadm contribution / work cycle looks like and start contributing to kubeadm!
20. Thank you! @luxas on Github @kubernetesonarm on Twitter lucas@luxaslabs.com