CKA Specialization Unit-1 || Architecture, Installation, Config

Task 1: Understand Kubernetes architecture

Objective: Identify control-plane components, worker components, and their responsibilities.

Show all nodes

kubectl get nodes -o wide

View system components

kubectl get pods -n kube-system -o wide

Inspect control-plane components

kubectl get pods -n kube-system | grep -E
‘apiserver|scheduler|controller|etcd’

Check API server health

kubectl get –raw=’/readyz?verbose’

View cluster information

kubectl cluster-info

Expected: All three nodes are Ready. API server, scheduler, controller-manager, and etcd are healthy.

Task 2: Cluster and node health checks

Objective: Learn to inspect node conditions, capacity, and cluster resources.

kubectl get nodes
kubectl describe node k8sw1

Check CPU and memory usage

kubectl top nodes
kubectl top pods -A

List cluster resources

kubectl get all -A

Show node resource allocation

kubectl describe nodes | grep -A 8 “Allocated resources”

Check recent events

kubectl get events -A –sort-by=.lastTimestamp

Exercise: Identify which worker has the highest CPU usage and check whether any pod is failing.

Task 3: Explore kubeadm and cluster configuration

Objective: Understand how kubeadm bootstraps Kubernetes clusters.

Run on the control-plane node:

Check kubeadm version

kubeadm version

Inspect kubeadm cluster configuration

kubectl -n kube-system get cm kubeadm-config -o yaml

Inspect kubelet configuration

kubectl -n kube-system get cm kubelet-config -o yaml

Check kubelet service

sudo systemctl status kubelet

Inspect static pod manifests

sudo ls -l /etc/kubernetes/manifests/

Inspect Kubernetes certificates

sudo kubeadm certs check-expiration

Exercise: Find the API server endpoint, Pod CIDR, and Kubernetes version.

Task 4: Kubernetes networking

Objective: Understand Pod networking, Services, DNS, and CNI.

Check Pod IP addresses

kubectl get pods -A -o wide

Check Services

kubectl get svc -A

Inspect Flannel CNI

kubectl get daemonset -n kube-flannel
kubectl get pods -n kube-flannel -o wide

Check CoreDNS

kubectl get pods -n kube-system -l k8s-app=kube-dns

Test cluster DNS

kubectl run dns-test
–image=busybox:1.36
–restart=Never
–command — sleep 3600

kubectl exec dns-test — nslookup kubernetes.default.svc.cluster.local

Cleanup

kubectl delete pod dns-test

Exercise: Verify that CoreDNS resolves Kubernetes Service names.

Task 5: Deploy and manage an application

Objective: Practice Deployments, replicas, scaling, and rollout management.

Create Deployment

kubectl create deployment web –image=nginx:stable

Scale to three replicas

kubectl scale deployment web –replicas=3

Expose application internally

kubectl expose deployment web
–port=80 –target-port=80

Inspect application

kubectl get deployments,pods,svc -o wide

Update container image

kubectl set image deployment/web nginx=nginx:alpine

Monitor rollout

kubectl rollout status deployment/web

Check rollout history

kubectl rollout history deployment/web

Roll back

kubectl rollout undo deployment/web

Exercise: Confirm that three replicas are running and distributed across available nodes.

Task 6: Node maintenance

Objective: Safely remove workloads from a worker node for maintenance.

Mark worker unschedulable

kubectl cordon k8sw1

Drain workloads

kubectl drain k8sw1
–ignore-daemonsets
–delete-emptydir-data

Check node state

kubectl get nodes

Re-enable scheduling

kubectl uncordon k8sw1

Verify node

kubectl describe node k8sw1

Important: drain can disrupt applications, and --delete-emptydir-data permits deletion of temporary Pod data. Practice only with disposable workloads.

Exercise: Explain the difference between cordon, drain, and uncordon.

Task 7: Monitoring and troubleshooting

Objective: Identify failed workloads and troubleshoot Kubernetes components.

Find pods not Running or Succeeded

kubectl get pods -A
–field-selector=status.phase!=Running,status.phase!=Succeeded

Check deployment health

kubectl get deployments -A

Check daemonsets

kubectl get daemonsets -A

Inspect a pod

kubectl describe pod -n

Inspect container logs

kubectl logs -n

Previous container logs

kubectl logs -n –previous

Inspect kubelet logs on a node

sudo journalctl -u kubelet -n 100 –no-pager

Exercise: Create a pod with an invalid image and diagnose the ImagePullBackOff error.

Task 8: RBAC and Service Accounts

Objective: Implement least-privilege API access.

Create namespace

kubectl create namespace cka-lab

Create ServiceAccount

kubectl create serviceaccount cka-reader -n cka-lab

Create Role

kubectl create role pod-reader
–verb=get,list,watch
–resource=pods
-n cka-lab

Bind Role to ServiceAccount

kubectl create rolebinding pod-reader-binding
–role=pod-reader
–serviceaccount=cka-lab:cka-reader
-n cka-lab

Test authorization

kubectl auth can-i list pods
–as=system:serviceaccount:cka-lab:cka-reader
-n cka-lab

Test denied action

kubectl auth can-i delete pods
–as=system:serviceaccount:cka-lab:cka-reader
-n cka-lab

Expected: Listing pods returns yes; deleting pods returns no.

Task 9: Kubernetes cluster upgrades

Objective: Understand the kubeadm upgrade workflow.

For a practice lab, begin with read-only checks:

Check current versions

kubectl version
kubeadm version
kubelet –version

Check node versions

kubectl get nodes -o wide

Inspect upgrade plan on control-plane

sudo kubeadm upgrade plan

Upgrade workflow to study:

Do not execute an upgrade without checking version compatibility, package repositories, and the official version-specific procedure.

Task 10: High availability and etcd

Objective: Understand etcd, control-plane redundancy, and recovery.

Check etcd pod

kubectl get pods -n kube-system | grep etcd

Inspect etcd logs

kubectl logs -n kube-system etcd-k8sm –tail=50

Inspect etcd static pod configuration

sudo cat /etc/kubernetes/manifests/etcd.yaml

Check control-plane nodes

kubectl get nodes
-l node-role.kubernetes.io/control-plane

Inspect API server endpoints

kubectl get endpoints kubernetes -n default

Exercise: Explain why a single control-plane cluster is not highly available and why a three-member etcd cluster can tolerate one member failure.

Kubernetes(k8S) Definitions & Acronyms

KUBERNETES (K8s) – DEFINITIONS AND ACRONYMS

Plain-text reference guide for Kubernetes administrators and CKA preparation.

  1. KUBERNETES FUNDAMENTALS

K8s – Kubernetes (K + 8 letters + s).
Cluster – Group of nodes managed by Kubernetes.
Node – Physical or virtual machine participating in a cluster.
Control Plane – Components responsible for managing cluster state.
Worker Node – Machine that runs application workloads.
Pod – Smallest deployable Kubernetes unit; one or more containers.
Container – Isolated application process environment.


Namespace – A Namespace is a Kubernetes resource that logically separates and organizes resources within a single cluster.

It provides:

  • Resource Isolation: Separates applications and workloads logically.
  • Resource Management: Controls CPU, memory, and other resource quotas.
  • Access Control: Supports namespace-specific RBAC permissions.
  • Organization: Groups resources by team, project, or environment.
  • Name Separation: Allows resources with the same name in different namespaces.

Use Cases: Development, testing, production, and multi-team environments.

Key Point: Namespaces provide logical separation, not complete security or network isolation by default.

Manifest – YAML or JSON definition of a Kubernetes resource.
Desired State – Target configuration Kubernetes attempts to maintain.
Reconciliation – Process of bringing actual state toward desired state.
kubectl – Command-line client for the Kubernetes API.
kubeconfig – Configuration for clusters, users, and contexts.
Context – Selected cluster, user, and optional namespace.
API – Application Programming Interface.
YAML – YAML Ain’t Markup Language.
JSON – JavaScript Object Notation.
CRI – Container Runtime Interface.
OCI – Open Container Initiative.
CNI – Container Network Interface.
CSI – Container Storage Interface.

  1. CONTROL PLANE AND NODE COMPONENTS

kube-apiserver – Exposes and validates the Kubernetes API.
etcd – Distributed key-value database storing cluster state.
kube-scheduler – Assigns unscheduled Pods to suitable nodes.
kube-controller-manager – Runs core reconciliation controllers.
cloud-controller-manager (CCM) – Integrates cloud provider infrastructure.
kubelet – Node agent that manages Pod execution.
kube-proxy – Implements Service networking on nodes (when deployed).
CRD – CustomResourceDefinition; adds a custom API resource type.
Leader Election – Selects an active leader among controller replicas.

  1. WORKLOADS AND CONTROLLERS

Deployment – A Deployment is a Kubernetes resource that manages application Pods through ReplicaSets.

It provides:

  • Scaling: Increase or decrease Pod replicas.
  • Self-Healing: Automatically replace failed Pods.
  • Rolling Updates: Update applications with minimal downtime.
  • Rollback: Restore previous application versions.
  • Desired State: Maintain the specified number of replicas.

Architecture: Deployment → ReplicaSet → Pods


ReplicaSet – A ReplicaSet is a Kubernetes resource that ensures the desired number of identical Pod replicas are running at all times.

It provides:

  • Pod Management: Creates and maintains Pods.
  • Self-Healing: Replaces failed or deleted Pods.
  • Scaling: Maintains the specified number of replicas.
  • High Availability: Helps keep applications running through multiple Pod replicas.

Architecture: Deployment → ReplicaSet → Pods

Key Point: ReplicaSets are typically managed by Deployments rather than directly.


StatefulSet – A StatefulSet is a Kubernetes resource that manages stateful applications requiring stable identities, persistent storage, and ordered deployment.

It provides:

  • Stable Identity: Each Pod gets a unique, persistent name.
  • Persistent Storage: Maintains dedicated storage for each Pod.
  • Ordered Deployment: Creates and removes Pods in a defined order.
  • Self-Healing: Replaces failed Pods while preserving their identity.
  • Stable Networking: Provides predictable Pod DNS identities.

Architecture: StatefulSet → Pods → Persistent Volumes (PVCs)

Use Cases: Databases (MySQL, PostgreSQL, MongoDB), Kafka, and other stateful applications.

Key Point: Unlike Deployments, StatefulSets maintain stable Pod identities and storage across Pod replacements.

DaemonSet -A DaemonSet is a Kubernetes resource that ensures a copy of a specific Pod runs on every eligible node (or selected nodes) in a cluster.

It provides:

  • Node-Level Deployment: Runs one Pod per eligible node.
  • Automatic Scheduling: Deploys Pods when new eligible nodes join.
  • Self-Healing: Replaces failed or deleted Pods.
  • Centralized Management: Manages node-level services across the cluster.

Architecture: DaemonSet → One Pod per Eligible Node

Use Cases: Log collection (Fluent Bit), monitoring (Prometheus Node Exporter), and networking (Flannel, Calico).

Key Point: Unlike Deployments, DaemonSets ensure node-level Pod coverage rather than maintaining a fixed number of replicas.

Job – Runs tasks to completion.
CronJob (CJ) – Creates Jobs on a schedule.
Init – Container Runs setup tasks before application containers start.
Sidecar – Container Helper container supporting an application container.
Ephemeral Container – Temporary container for interactive troubleshooting.
HPA – Horizontal Pod Autoscaler; adjusts replica count.
VPA – Vertical Pod Autoscaler; recommends/adjusts Pod resources.
PDB – Pod Disruption Budget; limits voluntary disruptions.
Rolling Update – Gradually replaces old Pods with new Pods.
Rollback – Returns a workload to an earlier revision.
Restart Policy – Rules governing container restarts within a Pod.

  1. NETWORKING AND SERVICE DISCOVERY

CNI – Container Network Interface.
CoreDNS – DNS server commonly used for Kubernetes discovery.
ClusterIP – Service type reachable within the cluster.

NodePort – A NodePort is a Kubernetes Service type that exposes an application externally through a specific port on every cluster node.

It provides:

  • External Access: Allows access using NodeIP:NodePort.
  • Port Allocation: Uses ports 30000–32767 by default.
  • Traffic Distribution: Routes incoming traffic to backend Pods.
  • Node-Level Access: Makes the Service accessible through any reachable node IP.

Architecture: Client → NodeIP:NodePort → Kubernetes Service → Pods

Example Command:

LoadBalancer – A LoadBalancer is a Kubernetes Service type that exposes an application externally and distributes incoming network traffic across its backend Pods.

It provides:

  • External Access: Exposes applications through an external IP address.
  • Traffic Distribution: Routes incoming traffic to available backend Pods.
  • High Availability: Distributes traffic across multiple Pod replicas.
  • Cloud Integration: Uses cloud provider load balancers or solutions like MetalLB.

Architecture: Client → External Load Balancer → Kubernetes Service → Pods

Example Command:

kubectl expose deployment nginx \
  --type=LoadBalancer \
  --port=80

Key Point: A LoadBalancer Service provides external access to applications. On bare-metal clusters, an additional load balancer implementation, such as MetalLB, is typically required.


ExternalName – Service type returning a DNS alias.
Ingress API – resource describing HTTP/HTTPS routing rules.
Ingress – Controller Software that implements Ingress rules.
Gateway API – APIs for flexible service traffic routing.
NetworkPolicy – Policy controlling selected Pod ingress/egress traffic.
Pod CIDR – IP address range used for Pods.
Service CIDR – IP address range used for Services.
eBPF – Extended Berkeley Packet Filter.
Flannel – Kubernetes CNI network plugin.
Calico – Kubernetes networking and network policy solution.
Cilium – eBPF-based networking, security, and observability solution.

  1. STORAGE

Volume – Storage exposed to containers in a Pod.
PV – PersistentVolume; cluster storage resource.
PVC – PersistentVolumeClaim; request for persistent storage.
SC – StorageClass; defines storage provisioning properties.
CSI – Container Storage Interface.
Storage Provisioner – Component that provisions storage volumes.
Dynamic Provisioning – Automatic creation of PVs for claims.
Static Provisioning – Pre-created PVs managed by administrators.
emptyDir – Ephemeral volume tied to a Pod’s lifetime.
hostPath – Mount of a node filesystem path into a Pod.
RWO – ReadWriteOnce; read/write mount by one node.
ROX – ReadOnlyMany; read-only mounts by many nodes.
RWX – ReadWriteMany; read/write mounts by many nodes.
RWOP – ReadWriteOncePod; exclusive read/write use by one Pod.
NFS – Network File System.
iSCSI – Internet Small Computer Systems Interface.
VolumeSnapshot – Snapshot resource supported by CSI snapshot components.

  1. SECURITY AND ACCESS CONTROL

RBAC – Role-Based Access Control.
SA – ServiceAccount; identity for workloads.
Role – Set of permissions within a namespace.
ClusterRole – Permissions that can apply cluster-wide or to namespaces.
RoleBinding – Grants Role or ClusterRole permissions in a namespace.
ClusterRoleBinding – Grants ClusterRole permissions cluster-wide.
TLS – Transport Layer Security.
mTLS – Mutual Transport Layer Security.
CA – Certificate Authority.
CSR – Certificate Signing Request.
OIDC – OpenID Connect.
JWT – JSON Web Token.
PSA – Pod Security Admission.
PSS – Pod Security Standards.
SCC – Security Context Constraints (OpenShift).
Secret – Resource storing sensitive configuration data.
ConfigMap (CM) – Resource storing non-confidential configuration data.
Admission Controller – Validates or mutates API requests.
SecurityContext -Pod/container security settings.

  1. SCHEDULING AND RESOURCE MANAGEMENT

CPU Request – CPU quantity used by the scheduler for placement.
CPU Limit Upper – CPU usage bound for a container.
Memory Request – Memory quantity used for scheduling.
Memory Limit – Upper memory bound for a container.
QoS – Quality of Service classification for Pods.
Guaranteed QoS – when CPU/memory requests equal limits for all containers.
Burstable QoS – with some requests/limits but not Guaranteed.
BestEffort QoS – without CPU/memory requests or limits.
Taint – Node restriction that repels non-tolerating Pods.
Toleration – Pod setting allowing scheduling past a matching taint.
NodeSelector – Label-based node selection.
Node Affinity – Flexible node scheduling constraints/preferences.
Pod Affinity – Places Pods near matching Pods.
Pod Anti-Affinity – Separates Pods from matching Pods.
Eviction – Removal of a Pod from a node.
OOMKilled – Container killed due to out-of-memory condition.
ResourceQuota – Limits aggregate resource usage in a namespace.
LimitRange – Defines defaults and bounds for resource requests/limits.
PriorityClass – Defines a Pod’s scheduling priority.

  1. OBSERVABILITY AND TROUBLESHOOTING

Metrics Server – Supplies resource metrics for kubectl top and HPA.
Prometheus – Metrics collection and time-series monitoring system.
Grafana – Dashboard and visualization platform.
kube-state-metrics (KSM) – Exposes Kubernetes object-state metrics.
Node Exporter – Exposes node OS and hardware metrics.
OpenTelemetry (OTel) – Framework for metrics, logs, and traces.
CRI-O – Kubernetes-compatible container runtime.
containerd – Container runtime commonly used with Kubernetes.
CrashLoopBackOff – Repeated container crashes with increasing restart delay.
ImagePullBackOff – Backoff after unsuccessful image pull attempts.
ErrImagePull – Image retrieval failed.
Pending Pod – accepted but not fully set up to run.
Running Pod – bound to a node; containers created, starting, or running.
Liveness – Probe Detects when a container should be restarted.
Readiness – Probe Controls whether a Pod is ready for Service traffic.
Startup Probe -Allows slow-starting applications to initialize.
Events Records – of notable resource-related occurrences.
kubectl logs – Displays container logs.
kubectl describe – Shows resource details and related events.
kubectl top – Displays resource usage when metrics are available.

  1. INSTALLATION AND LIFECYCLE

kubeadm – Tool for bootstrapping Kubernetes clusters.
kubeadm init – Initializes a control plane.
kubeadm join – Adds a node to an existing cluster.
kubeadm upgrade – Assists with upgrades of kubeadm-managed clusters.
Helm – Kubernetes package manager.
Chart – Package of Kubernetes resources for Helm.
Release – Installed instance of a Helm chart.
Kustomize – Tool for customizing Kubernetes YAML configuration.
GitOps – Git-driven declarative deployment/operations approach.
Argo CD – GitOps continuous delivery controller.
Flux – GitOps toolkit for Kubernetes.
Drain – Evicts eligible Pods for node maintenance.
Cordon – Marks a node unschedulable.
Uncordon – Marks a node schedulable again.

  1. ADVANCED CLOUD-NATIVE TERMS

CR – Custom Resource; instance of a custom API type.
Operator – Controller using Kubernetes APIs to manage applications.
API Aggregation – Adds extension API servers to Kubernetes API surface.
Webhook – HTTP callback used for admission or integration.
Service Mesh – Infrastructure for service-to-service traffic management.
Istio – Service mesh platform.
Envoy – High-performance network proxy.
gRPC – Remote Procedure Call framework.
HTTP/2 – Hypertext Transfer Protocol version 2.
OPA – Open Policy Agent.
Kyverno – Kubernetes-native policy engine.
KEDA Kubernetes Event-driven Autoscaling.
Karpenter – Kubernetes node provisioning/lifecycle system.
Cluster Autoscaler – Adjusts node group sizes to meet workload needs.
SLO – Service Level Objective.
SLI – Service Level Indicator.
SLA – Service Level Agreement.
RTO – Recovery Time Objective.
RPO – Recovery Point Objective.
DR – Disaster Recovery.

  1. COMMON KUBECTL RESOURCE SHORT NAMES

Pods po – kubectl get po
Deployments – deploy kubectl get deploy
Services svc – kubectl get svc
Namespaces ns – kubectl get ns
Nodes no – kubectl get no
ReplicaSets rs – kubectl get rs
DaemonSets ds – kubectl get ds
StatefulSets sts – kubectl get sts
ConfigMaps cm – kubectl get cm
ServiceAccounts sa – kubectl get sa
PersistentVolumes pv – kubectl get pv
PersistentVolumeClaims pvc – kubectl get pvc
StorageClasses sc – kubectl get sc
NetworkPolicies netpol – kubectl get netpol
HorizontalPodAutoscalers hpa – kubectl get hpa
PodDisruptionBudgets pdb – kubectl get pdb
CronJobs cj – kubectl get cj
Ingresses ing – kubectl get ing
Endpoints ep – kubectl get ep

  1. TOP 20 CKA STUDY AREAS

  1. Pods and Deployments
  2. ReplicaSets
  3. StatefulSets and DaemonSets
  4. Jobs and CronJobs
  5. Services and DNS
  6. NetworkPolicies
  7. PersistentVolumes and PVCs
  8. StorageClasses
  9. RBAC and ServiceAccounts
  10. ConfigMaps and Secrets
  11. Taints and tolerations
  12. Node affinity and Pod affinity
  13. Resource requests and limits
  14. Liveness, readiness, and startup probes
  15. etcd backup and restore
  16. kubeadm installation and upgrades
  17. Node drain, cordon, and uncordon
  18. Pod and node troubleshooting
  19. Helm and Kustomize
  20. Cluster networking and CNI troubleshooting

RECOMMENDED LEARNING ORDER

Fundamentals -> Architecture -> Workloads -> Networking -> Storage ->
Security -> Scheduling -> Troubleshooting -> Cluster Administration

REFERENCES

Kubernetes documentation: https://kubernetes.io/docs/home/
CKA certification: https://www.cncf.io/training/certification/cka/

END OF DOCUMENT

Cilium – CNI for Kubernetes

What is Cilium?
Cilium is an open-source CNI (Container Network Interface) for Kubernetes. It uses Linux eBPF technology to provide high-performance networking, security, load balancing, and observability.

Where is it used?
Cilium is commonly used in Kubernetes environments for Pod-to-Pod networking, NetworkPolicy enforcement, Service load balancing, multi-cluster connectivity, and network monitoring. It is especially useful in large-scale cloud-native environments where performance, security, and visibility are important.

Key features: eBPF-based networking, L3/L4/L7 security policies, kube-proxy replacement, encryption with WireGuard/IPsec, and network observability through Hubble.

Alternatives: Calico is a popular feature-rich CNI with strong network-policy capabilities; Flannel is a simpler option mainly focused on Pod networking; Antrea is a Kubernetes-native networking and security solution based on Open vSwitch; and Kube-router provides networking and policy with a relatively lightweight architecture.

In short:
Cilium = Kubernetes networking + security + observability, powered by eBPF.

Sunday Kubernetes learning plan

Based on the LFS258 topics, here is a 12-week Sunday Kubernetes Fundamentals learning plan. I’ve arranged it so each week builds on the previous one and includes theory, commands, a diagram topic, and a practical lab.

WeekMain TopicKey Things to LearnCommands / Lab Focus
1Kubernetes FoundationsContainers vs Kubernetes, cluster concepts, control plane vs worker nodeskubectl version, kubectl cluster-info, kubectl get nodes
2Kubernetes ArchitectureAPI Server, Scheduler, Controller Manager, etcd, kubelet, kube-proxykubectl get pods -A, kubectl describe node
3Cluster Installationkubeadm, kubelet, kubectl, container runtime, cluster bootstrapkubeadm init, kubeadm join, kubectl get nodes
4Pods & API ObjectsPods, YAML, namespaces, labels, annotationskubectl run, kubectl get pod, kubectl describe, kubectl apply -f
5Deployments & WorkloadsReplicaSets, Deployments, scaling, rolling updates, rollbackkubectl create deployment, kubectl scale, kubectl rollout
6Services & NetworkingClusterIP, NodePort, service discovery, DNS, pod networkingkubectl expose, kubectl get svc, kubectl get endpoints
7Ingress & GatewayIngress Controller, HTTP routing, Gateway APICreate an Ingress and route traffic to two Services
8Kubernetes StorageVolumes, PV, PVC, StorageClass, CSI, Ceph integrationkubectl get pv,pvc,sc, create PVC and mount into a Pod
9Helm & KustomizePackage management, Helm charts, templating, overlayshelm install, helm list, helm upgrade, kubectl apply -k
10Scheduling & SecurityScheduler, affinity, taints/tolerations, RBAC, ServiceAccountskubectl taint, kubectl auth can-i, kubectl create role
11Troubleshooting & LoggingLogs, events, failed Pods, networking, resource problemskubectl logs, kubectl events, kubectl exec, kubectl top
12HA + CKA ReviewHA control plane, etcd, CRDs, backup/recovery, exam-style troubleshootingMixed troubleshooting lab + CKA-style exercises

Week 1 — Kubernetes Foundations

Start with the overall picture:

Application → Container → Pod → Worker Node → Kubernetes Cluster

Cover why Kubernetes exists, what problems it solves, Kubernetes vs Docker/Podman, clusters, nodes, Pods, and the basic declarative model.

Practice:

kubectl version
kubectl cluster-info
kubectl get nodes
kubectl get nodes -o wide
kubectl get pods -A

Diagram: Kubernetes high-level architecture.

Lab: Connect kubectl to a cluster and identify the control-plane and worker nodes.


Week 2 — Kubernetes Architecture

Study the components in more detail:

              Kubernetes Cluster

          +----------------------+
          |    Control Plane     |
          |----------------------|
          | kube-apiserver       |
          | scheduler            |
          | controller-manager   |
          | etcd                 |
          +----------+-----------+
                     |
        -----------------------------
        |                           |
+-------v-------+           +-------v-------+
| Worker Node 1 |           | Worker Node 2 |
|---------------|           |---------------|
| kubelet       |           | kubelet       |
| kube-proxy    |           | kube-proxy    |
| containerd    |           | containerd    |
| Pods          |           | Pods          |
+---------------+           +---------------+

Useful commands:

kubectl get pods -n kube-system
kubectl describe node <node-name>
kubectl get componentstatuses
kubectl get --raw='/readyz?verbose'

Lab: Identify which Kubernetes component performs scheduling, stores cluster state, communicates with nodes, and maintains desired state.


Week 3 — Building a Cluster

Learn how the pieces are installed:

Linux → containerd → kubelet → kubeadm → Kubernetes

Core commands:

sudo kubeadm init

Configure kubectl:

mkdir -p $HOME/.kube
sudo cp /etc/kubernetes/admin.conf $HOME/.kube/config
sudo chown $(id -u):$(id -g) $HOME/.kube/config

Worker:

sudo kubeadm join <control-plane-ip>:6443 \
  --token <token> \
  --discovery-token-ca-cert-hash sha256:<hash>

Lab: Build a small cluster or walk through a kubeadm deployment.


Week 4 — Pods and Kubernetes API Objects

Learn the most important Kubernetes object first: the Pod.

Example:

apiVersion: v1
kind: Pod
metadata:
  name: nginx
  labels:
    app: web
spec:
  containers:
    - name: nginx
      image: nginx

Run:

kubectl apply -f nginx.yaml
kubectl get pods
kubectl get pods -o wide
kubectl describe pod nginx
kubectl logs nginx

Also cover:

kubectl get namespaces
kubectl get pods -n kube-system
kubectl get pods --show-labels

Diagram: Pod → Container relationship.

Lab: Create two Pods using YAML and labels.


Week 5 — Deployments, ReplicaSets and Workloads

The relationship to understand is:

Deployment
    |
    v
ReplicaSet
    |
    v
+-----+ +-----+ +-----+
| Pod | | Pod | | Pod |
+-----+ +-----+ +-----+

Commands:

kubectl create deployment nginx --image=nginx
kubectl get deployments
kubectl get rs
kubectl get pods

Scale:

kubectl scale deployment nginx --replicas=4

Upgrade:

kubectl set image deployment/nginx nginx=nginx:1.27

Check:

kubectl rollout status deployment/nginx
kubectl rollout history deployment/nginx

Rollback:

kubectl rollout undo deployment/nginx

Lab: Deploy nginx, scale it, upgrade it, then roll it back.


Week 6 — Kubernetes Networking & Services

This is one of the most important Kubernetes topics.

Understand:

          Service
       10.96.10.20
            |
    -----------------
    |       |       |
   Pod     Pod     Pod
10.1.1.2 10.1.2.3 10.1.3.4

Study:

Pod IP → Service → ClusterIP → NodePort → external access

Commands:

kubectl expose deployment nginx \
  --port=80 \
  --type=ClusterIP

kubectl get svc
kubectl describe svc nginx
kubectl get endpoints

NodePort:

kubectl expose deployment nginx \
  --name=nginx-nodeport \
  --port=80 \
  --type=NodePort

Lab: Access an application first through Pod IP, then ClusterIP, then NodePort.


Week 7 — Ingress and Gateway API

Architecture:

Internet
   |
   v
Ingress / Gateway
   |
   +----------------+
   |                |
   v                v
Service A        Service B
   |                |
 Pods             Pods

Learn host/path routing such as:

example.com/app1 → service-app1
example.com/app2 → service-app2

Commands:

kubectl get ingress
kubectl describe ingress

Lab: Deploy two applications and route /app1 and /app2 to different Services.


Week 8 — Storage, PV, PVC, CSI and Ceph

This week connects directly to the Ceph diagrams we discussed.

Understand:

Application Pod
      |
      v
     PVC
      |
      v
      PV
      |
      v
 StorageClass
      |
      v
  CSI Driver
      |
      v
     Ceph

Commands:

kubectl get pv
kubectl get pvc
kubectl get storageclass
kubectl describe pvc <pvc-name>

Example PVC:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: app-data
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 5Gi

Lab: Create a PVC, attach it to a Pod, write data, delete/recreate the Pod, and verify that the data remains.


Week 9 — Helm & Kustomize

Learn why manually maintaining dozens of YAML files becomes difficult.

Helm:

helm repo add bitnami https://charts.bitnami.com/bitnami
helm repo update

helm install my-nginx bitnami/nginx
helm list
helm upgrade my-nginx bitnami/nginx
helm uninstall my-nginx

Kustomize:

kubectl apply -k .

Concept:

Base YAML
   |
   +---- Dev Overlay
   |
   +---- Test Overlay
   |
   +---- Production Overlay

Lab: Install an application with Helm and modify another application using Kustomize.


Week 10 — Scheduling and Security

Scheduling:

Pod
 |
Scheduler
 |
 +---- Node A
 +---- Node B
 +---- Node C

Study:

requests/limits, nodeSelector, affinity, anti-affinity, taints and tolerations.

Commands:

kubectl describe node
kubectl get pods -o wide

kubectl taint nodes worker1 dedicated=test:NoSchedule

Security topics:

User → Role → RoleBinding → Resource

Commands:

kubectl auth can-i get pods
kubectl auth can-i delete pods

kubectl create serviceaccount app-user

Lab: Restrict a ServiceAccount so it can read Pods but cannot delete them.


Week 11 — Troubleshooting

This should be one of the most practical sessions.

Use a troubleshooting sequence:

Application issue
       |
       v
kubectl get pods
       |
       v
kubectl describe
       |
       v
kubectl logs
       |
       v
Events / Service / DNS
       |
       v
Node / kubelet

Core commands:

kubectl get pods -A
kubectl get events --sort-by=.metadata.creationTimestamp

kubectl describe pod <pod>
kubectl logs <pod>
kubectl logs <pod> --previous

kubectl exec -it <pod> -- /bin/sh

kubectl top pods
kubectl top nodes

Also practice:

journalctl -u kubelet

Lab: Intentionally create:

  • Wrong container image
  • Wrong Service selector
  • Incorrect port
  • Failed readiness probe
  • Unschedulable Pod

Then diagnose each one.


Week 12 — High Availability + CKA Review

Finish with production Kubernetes architecture:

             Load Balancer
                  |
        ---------------------
        |         |         |
       CP1       CP2       CP3
        |         |         |
        ------- etcd -------
                  |
       ---------------------
       |         |         |
     Worker1   Worker2   Worker3

Study:

HA control plane, etcd backup/restore, CRDs, node maintenance, application recovery, cluster troubleshooting.

Useful operations:

kubectl cordon worker1
kubectl drain worker1 --ignore-daemonsets
kubectl uncordon worker1

Review:

kubectl get nodes
kubectl get pods -A
kubectl get svc
kubectl get ingress
kubectl get pv,pvc
kubectl get roles,rolebindings

Finish with a practical challenge: deploy an application with Deployment → Service → Ingress → PVC, intentionally introduce two faults, then troubleshoot them.

Recommended Sunday session structure

Since you already have a regular Kubernetes & Cloud Native learning session, a compact format can work well:

10 min — Concept → 5 min — Architecture diagram → 10 min — commands/demo → 5 min — troubleshooting/question

For deeper personal study, spend another 60–90 minutes during the week reproducing the lab yourself. The biggest improvement will come from typing the commands rather than only reading them.

After these 12 weeks, the natural next step is CKA-focused practice, especially timed exercises around troubleshooting, networking, storage, scheduling, RBAC, cluster maintenance, and kubectl speed.

CEPH – An open source distributed storage platform

Ceph is not an acronym, so there is no official expansion like “C.E.P.H.”

The name Ceph comes from “cephalopod”—animals such as an octopus or squid. The idea fits Ceph because it can spread and manage data across many storage nodes, somewhat like many arms working together.

Kubernetes handles container orchestration, scheduling, networking, and application lifecycle management, while Ceph provides the storage layer. Ceph MON, MGR, and OSD components work together to maintain cluster health, manage the storage environment, replicate data, and provide resilient persistent storage to Kubernetes applications.

Ceph combines multiple storage servers into one distributed storage platform. It can provide RBD block storage for virtual machines and Kubernetes, CephFS for shared file storage, and RGW for S3-compatible object storage. Because data is distributed and replicated across multiple OSDs, Ceph can automatically recover from disk or server failures while keeping applications running.

Kubernetes and Ceph provide a scalable, highly available, cloud-native platform for running stateful applications such as databases, monitoring systems, logging platforms, and other enterprise workloads.

What is SD-WAN and Benefits

SD-WAN (Software-Defined Wide Area Network) is a modern approach to managing and optimizing wide area networks (WANs), allowing businesses to securely and efficiently connect remote offices, data centers, and cloud resources over the internet. Unlike traditional WANs, which rely on expensive, static MPLS (Multiprotocol Label Switching) circuits or leased lines, SD-WAN uses software to dynamically manage the traffic across multiple types of network connections, such as broadband internet, 4G/5G, MPLS, and other network types.

How SD-WAN Works:

Centralized Control Plane:

    • SD-WAN is built around a centralized control plane that manages the entire network’s policies and traffic routing.
    • This control plane is typically hosted in the cloud or on-premises, and it communicates with SD-WAN devices (also called edge devices or appliances) at branch offices, data centers, or remote sites.
    • The centralized control allows for real-time traffic management and decision-making, optimizing network performance across different types of connections.

    Decentralized Data Plane:

      • The data plane is made up of SD-WAN devices located at the edge of the network, such as branch routers, and it handles actual data forwarding and traffic routing.
      • These edge devices are responsible for securely transmitting data between remote sites, data centers, and cloud applications, based on the policies set by the control plane.

      Traffic Management and Routing:

        • SD-WAN uses intelligent path selection to route traffic over the most appropriate and cost-effective path in real-time. It can choose from multiple links (e.g., MPLS, broadband, LTE) based on:
          • Performance metrics: latency, jitter, packet loss, etc.
          • Application requirements: certain applications might need high bandwidth or low latency.
          • Policy-driven decisions: predefined rules about how specific types of traffic should be prioritized (e.g., voice or video traffic).

        Application-Aware Routing:

          • SD-WAN can distinguish between different types of applications and automatically route traffic based on business priorities.
          • For example, it can prioritize VoIP or video conferencing traffic over general web browsing traffic to ensure high-quality performance for critical applications.
          • It can also dynamically adjust traffic routes based on network conditions to maintain application performance.

          Security:

            • SD-WAN often includes integrated security features such as:
              • Encryption: All traffic between SD-WAN devices is encrypted, ensuring secure communication over potentially untrusted public networks (e.g., the internet).
              • Firewalling: Built-in firewall capabilities can prevent unauthorized access and attacks.
              • VPN (Virtual Private Network): Secure site-to-site connections can be established, leveraging IPsec or SSL VPNs.
              • Zero Trust Security: Many SD-WAN solutions implement Zero Trust principles, ensuring that security policies are enforced across the network regardless of location.

            Cloud Integration:

              • SD-WAN is well-suited for cloud-first or hybrid IT environments because it allows direct and optimized access to cloud applications (e.g., SaaS, IaaS, PaaS) without routing traffic through centralized data centers.
              • This reduces latency, improves application performance, and enhances user experience by enabling direct internet breakout from remote sites to cloud services.

              Simplified Management:

                • SD-WAN solutions are often managed through a centralized, web-based portal, providing administrators with visibility into the entire network.
                • The portal allows for easy configuration, monitoring, troubleshooting, and reporting across all remote sites and cloud applications.
                • Many SD-WAN platforms offer automation, allowing for the rapid deployment of new branch sites or network changes without requiring manual configuration at each site.

                Key Benefits of SD-WAN:

                Cost Efficiency:

                  • By leveraging lower-cost internet connections (such as broadband or LTE) alongside or in place of expensive MPLS links, organizations can reduce their WAN costs significantly.

                  Improved Performance:

                    • SD-WAN can provide better application performance by selecting the best path based on real-time network conditions, reducing bottlenecks and improving the user experience.

                    Scalability:

                      • SD-WAN networks are easier to scale as businesses grow. New sites can be added quickly without the need for complex configurations or additional hardware.

                      Flexibility:

                        • SD-WAN can support multiple types of connections (e.g., MPLS, broadband, LTE, 5G), making it adaptable to a wide range of network environments.

                        Security:

                          • SD-WAN provides built-in encryption and secure connections, reducing the need for separate security appliances.

                          Cloud Optimization:

                            • SD-WAN helps businesses securely and efficiently connect to cloud applications and services without backhauling traffic through a central data center.

                            Centralized Control and Visibility:

                              • The centralized control plane gives IT teams a unified view of the network, simplifying management and troubleshooting.

                              Use Cases for SD-WAN:

                              1. Branch Office Connectivity: Connecting multiple branch offices securely and efficiently, with optimized performance for cloud applications.
                              2. Cloud Transformation: Ensuring seamless, secure access to cloud resources and applications for remote and branch locations.
                              3. Business Continuity: Using multiple network links to ensure high availability and failover in case of a link or site failure.
                              4. Remote Worker Access: Extending SD-WAN benefits to remote workers by securely connecting them to corporate applications via the internet.

                              Conclusion:

                              SD-WAN is revolutionizing the way organizations manage their WANs by using software to dynamically manage traffic, optimize application performance, and reduce costs. It provides a more flexible, secure, and efficient solution compared to traditional WAN architectures, making it particularly well-suited for modern cloud-driven, distributed enterprise environments.

                              ব্যাংকক এয়ারপোর্টে এসে পুলিশের ইন্টারোগেশন

                              বাংলাদেশ থেকে ব্যাংকক এয়ারপোর্টে নামলাম। সাত ঘন্টার ট্রানজিট। দুপুরে ফ্লাইটে যা দিয়েছিলো তা হজম হয়ে গিয়েছে। আসার সময় বার্গার কিং এ ডাবল ফিস বার্গার খেয়েছিলাম। সুইডেনের আরলান্ডা এয়ারপোর্টের থেকেও দাম বেশী। মেজাজ খারাপ হল যখন তারা বার্গার কিং এর ভিআইপি ডিসকাউন্ট কার্ড চিনলো না। আমার আগের ইন্টারন্যাশনাল ফ্লাইটের অধিকাংশগুলোর অভিজ্ঞতাগুলো ছিল কোম্পানীর কাজে। যার ফলে খাবারের দাম নিয়ে কখনোই মাথা ঘামাতে হয় নাই। স্পেশ্যালী বাংলাট্রাকের ট্যুরগুলো। যেখানে থাকার থাকো, যা খাবার খাও, যা করার করো – সবই অ্যাপ্রুভড। মালিকপক্ষের খোলা আর এমন উদার মানসিকতার কারনে বাংলা ট্রাকের চাকুরি একটা লাইফটাইম এক্সপেরিয়েন্সও বটে। যারা করেছে একমাত্র তারাই জানে।

                              সূবর্নভুমি এয়ারপোর্টে খাবারের দোকান নেহায়তই কম না। কিন্তু কেন জানি ক্ষুধা থাকার পরও খাওয়ার ইচ্ছে জাগলো না। আবারো বার্গার কিং এ বসলাম – যাস্ট একটা বড় ফ্রেঞ্চফ্রাই আর পানি নিয়ে। এটা শেষ করে হাঁটাহাঁটি শুরু করলাম এয়ারপোর্ট ঘুরে দেখার জন্য। বেশ বড় এয়ারপোর্ট। ইলেক্ট্রনিক্সের দোকান থেকে কিছু কেনার চেষ্টা করলাম, দাম হিসেব করে কেনার ইচ্ছে হলো না। শুধু একটা ড্রাইড মিক্সড ফ্রুট প্যাক কিনলাম। এটারো দাম অযথা বেশি মনে হল।

                              তিন ঘন্টা ঘোরার পর গেট নম্বরের ঘোষনা পেলাম। আমারটা সি-১০, এক প্রান্তের শেষ মাথায়। সেদিকে হাঁটা দিলাম। জাবিনকে কল দিলাম। সে তার ল্যাবে কাজ করছে। জানালো – ছেলেরা স্কুলে, পাকিস্তানি বেবী সিটার মেয়েটি আসবে তারাজকে স্কুল আনা আর আরীবকে বাসায় স্কুল ট্যাক্সি থেকে রিসিভ করার জন্য। বাকি সব ইন ট্রাক এ আছে।

                              এর পর নাজমুল কল দিলো। যথারীতি আপকামিং টেকনোলজি নিয়ে বিশ বছর আগের মতো আলোচনা করতে গিয়ে ঘন্টার পর ঘন্টা কিভাবে কেটে গেল টের পেলাম না।

                              থাই এয়ারোওয়েজের ক্রুরা চলে এসেছে। ঘন্টা খানেক বাকী প্লেন ছাড়ার। বোর্ডিং শুরু হবে অচিরেই। এর মাঝে দুই এয়ারপোর্ট পুলিশ খুব হন্তদন্ত হয়ে আসলো বোর্ডিং কাউন্টারে। একজন হেড টাইপের ক্রুর সাথে জরুরি আলাপে লিপ্ত হলো। হেড ক্রু তার কথা শুনে উত্তেজিত হয়ে পড়লো। দুইজন পুলিশকে রুমের এক কর্নারে বসিয়ে কোন বিশেষ ব্যক্তিকে খোঁজা শুরু করলো। সে খুঁজছে বোর্ডিং পাস ৩৪এ ধারী যাত্রীকে। ভাবলাম, বড় কোন ক্রিমিনাল হবে হয়তো। পাঁচ মিনিট পর টের পেলাম সেই বিশেষ ব্যক্তিটি আমি। একটু ফ্লাসব্যাকে গেলাম – এয়ারপোর্টে ঢোকার পর বাথরুমে গেছি, সুন্দর করে ফ্লাস করেছি। একটা শুকনো ফলের প্যাকেট কিনেছি সেটার মেমো আছে। বার্গার কিং এ খাওয়ার পর টেবিলও পরিস্কার করেছি। এর বাইরে এক থাই মেয়ের দিকে ফ্রাকশন অফ সেকেন্ডের জন্য তাকিয়েছিলাম বটে, তবে মনে হয় সেটা ক্রাইম হবার মতো কিছুনা।

                              দুই অফিসারের সামনে বসে আছি। বিদেশ বিভুঁইয়ে এসে পুলিশের ইন্টারোগেশনে, টেনশন হবার কথা। তবে আমি আমার স্বভাববিরুদ্ধ টেনশন ফ্রি মুডে। আসলে বুঝতেছি না, সমস্যাটা কি।

                              এক অফিসার আমার পাসপোর্ট চাইলো। দিলাম। সে চার রকমের ইউভি, আল্টাভি স্কানার দিয়ে পাসপোর্ট চেক ব্যাস্ত হল। দেড় মিনিট পর আমিই কিউরিয়াস মুডে জানতে চাইলাম

                              – কিছু পাইলা? সন্দেহজনক?

                              সে বিস্মিত হয়ে আমার দিকে তাকালো। চেকিং বন্ধ করে শুকনোমুখে পারপোর্ট তার সহযোগীকে দিল। সহযোগী পাসপোর্টের পাতাগুলো চেক করে খুশি হয়ে উঠলো। আমাকে জিজ্ঞস করলো

                              – তুমি সুইডেন থেকে আসছো?

                              – হুম।

                              – তোমার তো ভিসা নাই পাসপোর্টে।

                              আমি অলমোষ্ট বেকুব হয়ে ওয়ালেট থেক ইউটি কার্ড তার হাতে ধরায়ে দিলাম। সে একপলক দেখেই তার সহযোগীর হাতে হস্তান্তর করলো। সহযোগী আবার তার যন্ত্রগুলো নিয়ে ব্যস্ত হয়ে পড়লো। প্রায় দুই মিনিট পর আমি আবারো জিজ্ঞেস করলাম

                              – কিছু পাইলা? সন্দেহজনক?

                              – সে শক্তমুখে জানতে চাইলো এই কার্ড আসল কি না?

                              আমি আরো শক্তমুখে জানালাম

                              – এটা সুইডিস সরকার আমাকে দিয়েছে। ওরা বলতে পারবে। আমার হাতে চেক করার মতো কোন যন্ত্র নাই।

                              সে কার্ড ফেরত দিয়ে আরেক জনের দিকে তাকালো। সে আবার তার দিকে তাকিয়ে। আমি দুজনের দিকে। থাই এয়ারলাইন্সের ভদ্রলোক আমাদের দিকে। সে বোর্ডিং শুরু করতে পারছে না। শেষমেষ আমিই বললাম

                              – তোমাদের সমস্যা কি? আমাকে বলো। আমিই হেল্প করার চেষ্টা করি।

                              কয়েক সেকেন্ড মুখ চাওয়া চাওয়ি করার পর লিডার টাইপের জন বলল

                              – তুমি চার দিনের জন্য বাংলাদেশে গেলা তাও আবার কোন লাগেজ ছাড়া। এটা সন্দেহজনক।

                              এবার আমি হিতাহিত জ্ঞান হারালাম। আশেপাশে দেয়াল না থাকায় শুধু কপালটা ঠুকতে পারলাম না। আমি পাল্টা প্রশ্ন করলাম

                              – সাথে লাগেজ নাই এটা সমস্য? তাছাড়া এটা আমার ট্রানজিট এয়ারপোর্ট। আরলান্ডা বা শাহজালাল এটা নিয়ে প্রশ্ন করলো না। তোমাদের মাথা ব্যথা কেন?

                              সে জানালো থাই এয়ারয়েজের কাছে ব্যপারটা সন্দেহজনক মনে হয়েছে। তারা এটা রেইজ করায় আমরা ইনভেষ্টিগেশনে এসেছি। মেজাজ খারপ হয়ে গেল। মোবাইলে ফেসবুক ক্রোমে ওপেন করে ছবিসহ লাষ্ট ষ্ট্যাটাসের ট্রান্সলেশন পড়তে দিলাম। পড়ার পর একটা গ্রুপ ছবি দেখিয়ে বললো

                              – এরা কারা?

                              – আমার বন্ধুরা।

                              – তারা কি করেন?

                              – সবাই সরকারী অতি উচ্চ পদস্থ কর্মকর্তা।

                              মনে হলো এবার বুঝছে। শুধুমাত্র ‘সরি’ বলে তারা বিদেয় নিলো। বোর্ডিং শুরু হলো। বিমানে উঠে বসলাম। বিমান ছাড়ার কয়েক মিনিট আগে আবারো থাই এয়ারোয়েজের সেই ভদ্রলোক।

                              – আমি সরি, আজ অনেক সিট ফাঁকা যাচ্ছে তাই তোমার সিটটা একটু চেঞ্জ করে দিয়েছি। তোমার নুতন সিটটা কি একটু দেখবা?

                              আমি বললাম দেখার দরকার নাই। চলো যাই। ব্যাগ নিয়ে সামনের দিকে এগোলাম। সিট দিয়েছে। বিজনেস ক্লাসে। উইন্ডো সিট। ডিসপ্লেতে দেখাচ্ছে এগারো ঘন্টা পঞ্চান্ন মিনিটের ফ্লাইট। সিটটা হেলিয়ে দিয়ে গা মেলিয়ে দিলাম। জীবনের প্রথম বিজনেস ক্লাস।

                              আরামে চোখ বুজে এলো। যখন চোখ মেললাম ডেষ্টিনেশন মাত্র চার হাজার কিলোমিটার বাঁকী। প্রায় সাড়ে সাত ঘন্টা ঘুমিয়েছি। আরো ঘন্টা খানেক পর কেবিন ক্রু নাস্তা দিয়ে গেল। সাথে আপেল অরেঞ্জ জুস দুইটাই দিয়ে গেছে। কে বলে টাকায় সুখ কেনা যায় না? শুধু টাকা কিনতে পয়সা লাগে।

                              ওহ, বাই দা ওয়ে। আমি নিজেও টাকায় সুখ কেনা যায় না বিশ্বাসী পাবলিকের দলে। এই বিশ্বাসেই বিশাল এক প্রশান্তি আছে। তাছাড়া, সুখ কেনার জন্য এতো খাটবে কে?